Files
ietf-draft-analyzer/workspace/drafts/gap-analysis/draft-nennemann-agent-behavioral-verification-00.html
Christian Nennemann 2506b6325a
Some checks failed
CI / test (3.11) (push) Failing after 1m37s
CI / test (3.12) (push) Failing after 57s
feat: add draft data, gap analysis report, and workspace config
2026-04-06 18:47:15 +02:00

2512 lines
96 KiB
HTML
Raw Permalink Blame History

This file contains invisible Unicode characters
This file contains invisible Unicode characters that are indistinguishable to humans but may be processed differently by a computer. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
<!DOCTYPE html>
<html lang="en" class="Internet-Draft">
<head>
<meta charset="utf-8">
<meta content="Common,Latin" name="scripts">
<meta content="initial-scale=1.0" name="viewport">
<title>Agent Behavioral Verification and Performance Benchmarking</title>
<meta content="Christian Nennemann" name="author">
<meta content="
This document defines protocols for runtime
verification that deployed AI agents behave
according to their declared policies. It also
specifies standardized metrics and a framework
for benchmarking agent performance across
implementations. Behavioral Evidence Tokens
(BETs) extend the Execution Context Token
architecture to provide cryptographically
verifiable proof of policy compliance.
Performance profiles enable objective comparison
of agent capabilities.
" name="description">
<meta content="xml2rfc 3.31.0" name="generator">
<meta content="draft-nennemann-agent-behavioral-verification-00" name="ietf.draft">
<!-- Generator version information:
xml2rfc 3.31.0
Python 3.14.3
ConfigArgParse 1.7.1
google-i18n-address 3.1.1
intervaltree 3.2.1
Jinja2 3.1.6
lxml 6.0.2
platformdirs 4.9.2
pycountry 26.2.16
PyYAML 6.0.3
requests 2.32.5
wcwidth 0.6.0
-->
<link href="draft-nennemann-agent-behavioral-verification-00.xml" rel="alternate" type="application/rfc+xml">
<link href="#copyright" rel="license">
<style type="text/css">/*
NOTE: Changes at the bottom of this file overrides some earlier settings.
Once the style has stabilized and has been adopted as an official RFC style,
this can be consolidated so that style settings occur only in one place, but
for now the contents of this file consists first of the initial CSS work as
provided to the RFC Formatter (xml2rfc) work, followed by itemized and
commented changes found necessary during the development of the v3
formatters.
*/
/* fonts */
@import url('https://static.ietf.org/fonts/noto-sans/import.css'); /* Sans-serif */
@import url('https://static.ietf.org/fonts/noto-serif/import.css'); /* Serif (print) */
@import url('https://static.ietf.org/fonts/roboto-mono/import.css'); /* Monospace */
:root {
--font-sans: 'Noto Sans', Arial, Helvetica, sans-serif;
--font-serif: 'Noto Serif', 'Times', 'Times New Roman', serif;
--font-mono: 'Roboto Mono', Courier, 'Courier New', monospace;
}
@viewport {
zoom: 1.0;
}
@-ms-viewport {
width: extend-to-zoom;
zoom: 1.0;
}
/* general and mobile first */
html {
}
body {
max-width: 90%;
margin: 1.5em auto;
color: #222;
background-color: #fff;
font-size: 14px;
font-family: var(--font-sans);
line-height: 1.6;
scroll-behavior: smooth;
overflow-wrap: break-word;
}
.ears {
display: none;
}
/* headings */
#title, h1, h2, h3, h4, h5, h6 {
margin: 1em 0 0.5em;
font-weight: bold;
line-height: 1.3;
}
#title {
clear: both;
border-bottom: 1px solid #ddd;
margin: 0 0 0.5em 0;
padding: 1em 0 0.5em;
}
.author {
padding-bottom: 4px;
}
h1 {
font-size: 26px;
margin: 1em 0;
}
h2 {
font-size: 22px;
margin-top: -20px; /* provide offset for in-page anchors */
padding-top: 33px;
}
h3 {
font-size: 18px;
margin-top: -36px; /* provide offset for in-page anchors */
padding-top: 42px;
}
h4 {
font-size: 16px;
margin-top: -36px; /* provide offset for in-page anchors */
padding-top: 42px;
}
h5, h6 {
font-size: 14px;
}
#n-copyright-notice {
border-bottom: 1px solid #ddd;
padding-bottom: 1em;
margin-bottom: 1em;
}
/* general structure */
p {
padding: 0;
margin: 0 0 1em 0;
text-align: left;
}
div, span {
position: relative;
}
div {
margin: 0;
}
.alignRight.art-text {
background-color: #f9f9f9;
border: 1px solid #eee;
border-radius: 3px;
padding: 1em 1em 0;
margin-bottom: 1.5em;
}
.alignRight.art-text pre {
padding: 0;
}
.alignRight {
margin: 1em 0;
}
.alignRight > *:first-child {
border: none;
margin: 0;
float: right;
clear: both;
}
.alignRight > *:nth-child(2) {
clear: both;
display: block;
border: none;
}
svg {
display: block;
}
@media print {
svg {
max-height: 850px;
max-width: 660px;
}
}
svg[font-family~="serif" i], svg [font-family~="serif" i] {
font-family: var(--font-serif);
}
svg[font-family~="sans-serif" i], svg [font-family~="sans-serif" i] {
font-family: var(--font-sans);
}
svg[font-family~="monospace" i], svg [font-family~="monospace" i] {
font-family: var(--font-mono);
}
.alignCenter.art-text {
background-color: #f9f9f9;
border: 1px solid #eee;
border-radius: 3px;
padding: 1em 1em 0;
margin-bottom: 1.5em;
}
.alignCenter.art-text pre {
padding: 0;
}
.alignCenter {
margin: 1em 0;
}
.alignCenter > *:first-child {
display: table;
border: none;
margin: 0 auto;
}
/* lists */
ol, ul {
padding: 0;
margin: 0 0 1em 2em;
}
ol ol, ul ul, ol ul, ul ol {
margin-left: 1em;
}
li {
margin: 0 0 0.25em 0;
}
.ulCompact li {
margin: 0;
}
ul.empty, .ulEmpty {
list-style-type: none;
}
ul.empty li, .ulEmpty li {
margin-top: 0.5em;
}
ul.ulBare, li.ulBare {
margin-left: 0em !important;
}
ul.compact, .ulCompact,
ol.compact, .olCompact {
line-height: 100%;
margin: 0 0 0 2em;
}
/* definition lists */
dl {
}
dl > dt {
float: left;
margin-right: 1em;
}
/*
dl.nohang > dt {
float: none;
}
*/
dl > dd {
margin-bottom: .8em;
min-height: 1.3em;
}
dl.compact > dd, .dlCompact > dd {
margin-bottom: 0em;
}
dl > dd > dl {
margin-top: 0.5em;
margin-bottom: 0em;
}
/* links */
a {
text-decoration: none;
}
a[href] {
color: #22e; /* Arlen: WCAG 2019 */
}
a[href]:hover {
background-color: #f2f2f2;
}
figcaption a[href],
a[href].selfRef {
color: #222;
}
/* XXX probably not this:
a.selfRef:hover {
background-color: transparent;
cursor: default;
} */
/* Figures */
tt, code, pre {
background-color: #f9f9f9;
font-family: var(--font-mono);
}
pre {
border: 1px solid #eee;
margin: 0;
padding: 1em;
}
img {
max-width: 100%;
}
figure {
margin: 0;
}
figure blockquote {
margin: 0.8em 0.4em 0.4em;
}
figcaption {
font-style: italic;
margin: 0 0 1em 0;
}
@media screen {
pre {
overflow-x: auto;
max-width: 100%;
max-width: calc(100% - 22px);
}
}
/* aside, blockquote */
aside, blockquote {
margin-left: 0;
padding: 1.2em 2em;
}
blockquote {
background-color: #f9f9f9;
color: #111; /* Arlen: WCAG 2019 */
border: 1px solid #ddd;
border-radius: 3px;
margin: 1em 0;
}
blockquote > *:last-child {
margin-bottom: 0;
}
cite {
display: block;
text-align: right;
font-style: italic;
}
.xref {
overflow-wrap: normal;
}
/* tables */
table {
width: 100%;
margin: 0 0 1em;
border-collapse: collapse;
border: 1px solid #eee;
}
th, td {
text-align: left;
vertical-align: top;
padding: 0.5em 0.75em;
}
th {
text-align: left;
background-color: #e9e9e9;
}
tr:nth-child(2n+1) > td {
background-color: #f5f5f5;
}
table caption {
font-style: italic;
margin: 0;
padding: 0;
text-align: left;
}
table p {
/* XXX to avoid bottom margin on table row signifiers. If paragraphs should
be allowed within tables more generally, it would be far better to select on a class. */
margin: 0;
}
/* pilcrow */
a.pilcrow {
color: #666; /* Arlen: AHDJ 2019 */
text-decoration: none;
visibility: hidden;
user-select: none;
-ms-user-select: none;
-o-user-select:none;
-moz-user-select: none;
-khtml-user-select: none;
-webkit-user-select: none;
-webkit-touch-callout: none;
}
@media screen {
aside:hover > a.pilcrow,
p:hover > a.pilcrow,
blockquote:hover > a.pilcrow,
div:hover > a.pilcrow,
li:hover > a.pilcrow,
pre:hover > a.pilcrow {
visibility: visible;
}
a.pilcrow:hover {
background-color: transparent;
}
}
/* misc */
hr {
border: 0;
border-top: 1px solid #eee;
}
.bcp14 {
font-variant: small-caps;
}
.role {
font-variant: all-small-caps;
}
/* info block */
#identifiers {
margin: 0;
font-size: 0.9em;
}
#identifiers dt {
width: 3em;
clear: left;
}
#identifiers dd {
float: left;
margin-bottom: 0;
}
/* Fix PDF info block run off issue */
@media print {
#identifiers dd {
max-width: 100%;
}
}
#identifiers .authors .author {
display: inline-block;
margin-right: 1.5em;
}
#identifiers .authors .org {
font-style: italic;
}
/* The prepared/rendered info at the very bottom of the page */
.docInfo {
color: #666; /* Arlen: WCAG 2019 */
font-size: 0.9em;
font-style: italic;
margin-top: 2em;
}
.docInfo .prepared {
float: left;
}
.docInfo .prepared {
float: right;
}
/* table of contents */
#toc {
padding: 0.75em 0 2em 0;
margin-bottom: 1em;
}
nav.toc ul {
margin: 0 0.5em 0 0;
padding: 0;
list-style: none;
}
nav.toc li {
line-height: 1.3em;
margin: 0.75em 0;
padding-left: 1.2em;
text-indent: -1.2em;
}
/* references */
.references dt {
text-align: right;
font-weight: bold;
min-width: 7em;
}
.references dd {
margin-left: 8em;
overflow: auto;
}
.refInstance {
margin-bottom: 1.25em;
}
.refSubseries {
margin-bottom: 1.25em;
}
.references .ascii {
margin-bottom: 0.25em;
}
/* index */
.index ul {
margin: 0 0 0 1em;
padding: 0;
list-style: none;
}
.index ul ul {
margin: 0;
}
.index li {
margin: 0;
text-indent: -2em;
padding-left: 2em;
padding-bottom: 5px;
}
.indexIndex {
margin: 0.5em 0 1em;
}
.index a {
font-weight: 700;
}
/* make the index two-column on all but the smallest screens */
@media (min-width: 600px) {
.index ul {
-moz-column-count: 2;
-moz-column-gap: 20px;
}
.index ul ul {
-moz-column-count: 1;
-moz-column-gap: 0;
}
}
/* authors */
address.vcard {
font-style: normal;
margin: 1em 0;
}
address.vcard .nameRole {
font-weight: 700;
margin-left: 0;
}
address.vcard .label {
font-family: var(--font-sans);
margin: 0.5em 0;
}
address.vcard .type {
display: none;
}
.alternative-contact {
margin: 1.5em 0 1em;
}
hr.addr {
border-top: 1px dashed;
margin: 0;
color: #ddd;
max-width: calc(100% - 16px);
}
/* temporary notes */
.rfcEditorRemove::before {
position: absolute;
top: 0.2em;
right: 0.2em;
padding: 0.2em;
content: "The RFC Editor will remove this note";
color: #9e2a00; /* Arlen: WCAG 2019 */
background-color: #ffd; /* Arlen: WCAG 2019 */
}
.rfcEditorRemove {
position: relative;
padding-top: 1.8em;
background-color: #ffd; /* Arlen: WCAG 2019 */
border-radius: 3px;
}
.cref {
background-color: #ffd; /* Arlen: WCAG 2019 */
padding: 2px 4px;
}
.crefSource {
font-style: italic;
}
/* alternative layout for smaller screens */
@media screen and (max-width: 1023px) {
body {
padding-top: 2em;
}
#title {
padding: 1em 0;
}
h1 {
font-size: 24px;
}
h2 {
font-size: 20px;
margin-top: -18px; /* provide offset for in-page anchors */
padding-top: 38px;
}
#identifiers dd {
max-width: 60%;
}
#toc {
position: fixed;
z-index: 2;
top: 0;
right: 0;
padding: 0;
margin: 0;
background-color: inherit;
border-bottom: 1px solid #ccc;
}
#toc h2 {
margin: -1px 0 0 0;
padding: 4px 0 4px 6px;
padding-right: 1em;
min-width: 190px;
font-size: 1.1em;
text-align: right;
background-color: #444;
color: white;
cursor: pointer;
}
#toc h2::before { /* css hamburger */
float: right;
position: relative;
width: 1em;
height: 1px;
left: -164px;
margin: 6px 0 0 0;
background: white none repeat scroll 0 0;
box-shadow: 0 4px 0 0 white, 0 8px 0 0 white;
content: "";
}
#toc nav {
display: none;
padding: 0.5em 1em 1em;
overflow: auto;
height: calc(100vh - 48px);
border-left: 1px solid #ddd;
}
}
/* alternative layout for wide screens */
@media screen and (min-width: 1024px) {
body {
max-width: 724px;
margin: 42px auto;
padding-left: 1.5em;
padding-right: 29em;
}
#toc {
position: fixed;
top: 42px;
right: 42px;
width: 25%;
margin: 0;
padding: 0 1em;
z-index: 1;
}
#toc h2 {
border-top: none;
border-bottom: 1px solid #ddd;
font-size: 1em;
font-weight: normal;
margin: 0;
padding: 0.25em 1em 1em 0;
}
#toc nav {
display: block;
height: calc(90vh - 84px);
bottom: 0;
padding: 0.5em 0 0;
overflow: auto;
}
img { /* future proofing */
max-width: 100%;
height: auto;
}
}
/* pagination */
@media print {
body {
width: 100%;
}
p {
orphans: 3;
widows: 3;
}
#n-copyright-notice {
border-bottom: none;
}
#toc, #n-introduction {
page-break-before: always;
}
#toc {
border-top: none;
padding-top: 0;
}
figure, pre {
page-break-inside: avoid;
}
figure {
overflow: scroll;
}
.breakable pre {
break-inside: auto;
}
h1, h2, h3, h4, h5, h6 {
page-break-after: avoid;
}
h2+*, h3+*, h4+*, h5+*, h6+* {
page-break-before: avoid;
}
pre {
white-space: pre-wrap;
word-wrap: break-word;
font-size: 10pt;
}
table {
border: 1px solid #ddd;
}
td {
border-top: 1px solid #ddd;
}
}
/* This is commented out here, as the string-set: doesn't
pass W3C validation currently */
/*
.ears thead .left {
string-set: ears-top-left content();
}
.ears thead .center {
string-set: ears-top-center content();
}
.ears thead .right {
string-set: ears-top-right content();
}
.ears tfoot .left {
string-set: ears-bottom-left content();
}
.ears tfoot .center {
string-set: ears-bottom-center content();
}
.ears tfoot .right {
string-set: ears-bottom-right content();
}
*/
@page :first {
padding-top: 0;
@top-left {
content: normal;
border: none;
}
@top-center {
content: normal;
border: none;
}
@top-right {
content: normal;
border: none;
}
}
@page {
size: A4;
margin-bottom: 45mm;
padding-top: 20px;
/* The following is commented out here, but set appropriately by in code, as
the content depends on the document */
/*
@top-left {
content: 'Internet-Draft';
vertical-align: bottom;
border-bottom: solid 1px #ccc;
}
@top-left {
content: string(ears-top-left);
vertical-align: bottom;
border-bottom: solid 1px #ccc;
}
@top-center {
content: string(ears-top-center);
vertical-align: bottom;
border-bottom: solid 1px #ccc;
}
@top-right {
content: string(ears-top-right);
vertical-align: bottom;
border-bottom: solid 1px #ccc;
}
@bottom-left {
content: string(ears-bottom-left);
vertical-align: top;
border-top: solid 1px #ccc;
}
@bottom-center {
content: string(ears-bottom-center);
vertical-align: top;
border-top: solid 1px #ccc;
}
@bottom-right {
content: '[Page ' counter(page) ']';
vertical-align: top;
border-top: solid 1px #ccc;
}
*/
}
/* Changes introduced to fix issues found during implementation */
/* Make sure links are clickable even if overlapped by following H* */
a {
z-index: 2;
}
/* Separate body from document info even without intervening H1 */
section {
clear: both;
}
/* Top align author divs, to avoid names without organization dropping level with org names */
.author {
vertical-align: top;
}
/* Leave room in document info to show Internet-Draft on one line */
#identifiers dt {
width: 8em;
}
/* Don't waste quite as much whitespace between label and value in doc info */
#identifiers dd {
margin-left: 1em;
}
/* Give floating toc a background color (needed when it's a div inside section */
#toc {
background-color: white;
}
/* Make the collapsed ToC header render white on gray also when it's a link */
@media screen and (max-width: 1023px) {
#toc h2 a,
#toc h2 a:link,
#toc h2 a:focus,
#toc h2 a:hover,
#toc a.toplink,
#toc a.toplink:hover {
color: white;
background-color: #444;
text-decoration: none;
}
}
/* Give the bottom of the ToC some whitespace */
@media screen and (min-width: 1024px) {
#toc {
padding: 0 0 1em 1em;
}
}
/* Style section numbers with more space between number and title */
.section-number {
padding-right: 0.5em;
}
/* prevent monospace from becoming overly large */
tt, code, pre {
font-size: 95%;
}
/* Fix the height/width aspect for ascii art*/
.sourcecode pre,
.art-text pre {
line-height: 1.12;
}
/* Add styling for a link in the ToC that points to the top of the document */
a.toplink {
float: right;
margin-right: 0.5em;
}
/* Fix the dl styling to match the RFC 7992 attributes */
dl > dt,
dl.dlParallel > dt {
float: left;
margin-right: 1em;
}
dl.dlNewline > dt {
float: none;
}
/* Provide styling for table cell text alignment */
table td.text-left,
table th.text-left {
text-align: left;
}
table td.text-center,
table th.text-center {
text-align: center;
}
table td.text-right,
table th.text-right {
text-align: right;
}
/* Make the alternative author contact information look less like just another
author, and group it closer with the primary author contact information */
.alternative-contact {
margin: 0.5em 0 0.25em 0;
}
address .non-ascii {
margin: 0 0 0 2em;
}
/* With it being possible to set tables with alignment
left, center, and right, { width: 100%; } does not make sense */
table {
width: auto;
}
/* Avoid reference text that sits in a block with very wide left margin,
because of a long floating dt label.*/
.references dd {
overflow: visible;
}
/* Control caption placement */
caption {
caption-side: bottom;
}
/* Limit the width of the author address vcard, so names in right-to-left
script don't end up on the other side of the page. */
address.vcard {
max-width: 30em;
margin-right: auto;
}
/* For address alignment dependent on LTR or RTL scripts */
address div.left {
text-align: left;
}
address div.right {
text-align: right;
}
/* Provide table alignment support. We can't use the alignX classes above
since they do unwanted things with caption and other styling. */
table.right {
margin-left: auto;
margin-right: 0;
}
table.center {
margin-left: auto;
margin-right: auto;
}
table.left {
margin-left: 0;
margin-right: auto;
}
/* Give the table caption label the same styling as the figcaption */
caption a[href] {
color: #222;
}
@media print {
.toplink {
display: none;
}
/* avoid overwriting the top border line with the ToC header */
#toc {
padding-top: 1px;
}
/* Avoid page breaks inside dl and author address entries */
.vcard {
page-break-inside: avoid;
}
}
/* Tweak the bcp14 keyword presentation */
.bcp14 {
font-variant: small-caps;
font-weight: bold;
font-size: 0.9em;
}
/* Tweak the invisible space above H* in order not to overlay links in text above */
h2 {
margin-top: -18px; /* provide offset for in-page anchors */
padding-top: 31px;
}
h3 {
margin-top: -18px; /* provide offset for in-page anchors */
padding-top: 24px;
}
h4 {
margin-top: -18px; /* provide offset for in-page anchors */
padding-top: 24px;
}
/* Float artwork pilcrow to the right */
@media screen {
.artwork a.pilcrow {
display: block;
line-height: 0.7;
margin-top: 0.15em;
}
}
/* Make pilcrows on dd visible */
@media screen {
dd:hover > a.pilcrow {
visibility: visible;
}
}
/* Make the placement of figcaption match that of a table's caption
by removing the figure's added bottom margin */
.alignLeft.art-text,
.alignCenter.art-text,
.alignRight.art-text {
margin-bottom: 0;
}
.alignLeft,
.alignCenter,
.alignRight {
margin: 1em 0 0 0;
}
/* In print, the pilcrow won't show on hover, so prevent it from taking up space,
possibly even requiring a new line */
@media print {
a.pilcrow {
display: none;
}
}
/* Styling for the external metadata */
div#external-metadata {
background-color: #eee;
padding: 0.5em;
margin-bottom: 0.5em;
display: none;
}
div#internal-metadata {
padding: 0.5em; /* to match the external-metadata padding */
}
/* Styling for title RFC Number */
h1#rfcnum {
clear: both;
margin: 0 0 -1em;
padding: 1em 0 0 0;
}
/* Make .olPercent look the same as <ol><li> */
dl.olPercent > dd {
margin-bottom: 0.25em;
min-height: initial;
}
/* Give aside some styling to set it apart */
aside {
border-left: 1px solid #ddd;
margin: 1em 0 1em 2em;
padding: 0.2em 2em;
}
aside > dl,
aside > ol,
aside > ul,
aside > table,
aside > p {
margin-bottom: 0.5em;
}
/* Additional page break settings */
@media print {
figcaption, table caption {
page-break-before: avoid;
}
}
/* Font size adjustments for print */
@media print {
body { font-size: 10pt; line-height: normal; max-width: 96%; }
h1 { font-size: 1.72em; padding-top: 1.5em; } /* 1*1.2*1.2*1.2 */
h2 { font-size: 1.44em; padding-top: 1.5em; } /* 1*1.2*1.2 */
h3 { font-size: 1.2em; padding-top: 1.5em; } /* 1*1.2 */
h4 { font-size: 1em; padding-top: 1.5em; }
h5, h6 { font-size: 1em; margin: initial; padding: 0.5em 0 0.3em; }
}
/* Sourcecode margin in print, when there's no pilcrow */
@media print {
.artwork,
.artwork > pre,
.sourcecode {
margin-bottom: 1em;
}
}
/* Avoid narrow tables forcing too narrow table captions, which may render badly */
table {
min-width: 20em;
}
/* ol type a */
ol.type-a { list-style-type: lower-alpha; }
ol.type-A { list-style-type: upper-alpha; }
ol.type-i { list-style-type: lower-roman; }
ol.type-I { list-style-type: upper-roman; }
/* Apply the print table and row borders in general, on request from the RPC,
and increase the contrast between border and odd row background slightly */
table {
border: 1px solid #ddd;
}
td {
border-top: 1px solid #ddd;
}
tr {
break-inside: avoid;
}
tr:nth-child(2n+1) > td {
background-color: #f8f8f8;
}
/* Use style rules to govern display of the TOC. */
@media screen and (max-width: 1023px) {
#toc nav { display: none; }
#toc.active nav { display: block; }
}
/* Add support for keepWithNext */
.keepWithNext {
break-after: avoid-page;
break-after: avoid-page;
}
/* Add support for keepWithPrevious */
.keepWithPrevious {
break-before: avoid-page;
}
/* Change the approach to avoiding breaks inside artwork etc. */
figure, pre, table, .artwork, .sourcecode {
break-before: auto;
break-after: auto;
}
/* Avoid breaks between <dt> and <dd> */
dl {
break-before: auto;
break-inside: auto;
}
dt {
break-before: auto;
break-after: avoid-page;
}
dd {
break-before: avoid-page;
break-after: auto;
orphans: 3;
widows: 3
}
span.break, dd.break {
margin-bottom: 0;
min-height: 0;
break-before: auto;
break-inside: auto;
break-after: auto;
}
/* Undo break-before ToC */
@media print {
#toc {
break-before: auto;
}
}
/* Text in compact lists should not get extra bottom margin space,
since that would makes the list not compact */
ul.compact p, .ulCompact p,
ol.compact p, .olCompact p {
margin: 0;
}
/* But the list as a whole needs the extra space at the end */
section ul.compact,
section .ulCompact,
section ol.compact,
section .olCompact {
margin-bottom: 1em; /* same as p not within ul.compact etc. */
}
/* The tt and code background above interferes with for instance table cell
backgrounds. Changed to something a bit more selective. */
tt, code {
background-color: transparent;
}
p tt, p code, li tt, li code, dt tt, dt code {
background-color: #f8f8f8;
}
/* Tweak the pre margin -- 0px doesn't come out well */
pre {
margin-top: 0.5px;
}
/* Tweak the compact list text */
ul.compact, .ulCompact,
ol.compact, .olCompact,
dl.compact, .dlCompact {
line-height: normal;
}
/* Don't add top margin for nested lists */
li > ul, li > ol, li > dl,
dd > ul, dd > ol, dd > dl,
dl > dd > dl {
margin-top: initial;
}
/* Elements that should not be rendered on the same line as a <dt> */
/* This should match the element list in writer.text.TextWriter.render_dl() */
dd > div.artwork:first-child,
dd > aside:first-child,
dd > blockquote:first-child,
dd > figure:first-child,
dd > ol:first-child,
dd > div.sourcecode:first-child,
dd > table:first-child,
dd > ul:first-child {
clear: left;
}
/* fix for weird browser behaviour when <dd/> is empty */
dt+dd:empty::before{
content: "\00a0";
}
/* Make paragraph spacing inside <li> smaller than in body text, to fit better within the list */
li > p {
margin-bottom: 0.5em
}
/* Don't let p margin spill out from inside list items */
li > p:last-of-type:only-child {
margin-bottom: 0;
}
</style>
<link href="rfc-local.css" rel="stylesheet" type="text/css">
<script type="application/javascript">async function addMetadata(){try{const e=document.styleSheets[0].cssRules;for(let t=0;t<e.length;t++)if(/#identifiers/.exec(e[t].selectorText)){const a=e[t].cssText.replace("#identifiers","#external-updates");document.styleSheets[0].insertRule(a,document.styleSheets[0].cssRules.length)}}catch(e){console.log(e)}const e=document.getElementById("external-metadata");if(e)try{var t,a="",o=function(e){const t=document.getElementsByTagName("meta");for(let a=0;a<t.length;a++)if(t[a].getAttribute("name")===e)return t[a].getAttribute("content");return""}("rfc.number");if(o){t="https://www.rfc-editor.org/rfc/rfc"+o+".json";try{const e=await fetch(t);a=await e.json()}catch(e){t=document.URL.indexOf("html")>=0?document.URL.replace(/html$/,"json"):document.URL+".json";const o=await fetch(t);a=await o.json()}}if(!a)return;e.style.display="block";const s="",d="https://datatracker.ietf.org/doc",n="https://datatracker.ietf.org/ipr/search",c="https://www.rfc-editor.org/info",l=a.doc_id.toLowerCase(),i=a.doc_id.slice(0,3).toLowerCase(),f=a.doc_id.slice(3).replace(/^0+/,""),u={status:"Status",obsoletes:"Obsoletes",obsoleted_by:"Obsoleted By",updates:"Updates",updated_by:"Updated By",see_also:"See Also",errata_url:"Errata"};let h="<dl style='overflow:hidden' id='external-updates'>";["status","obsoletes","obsoleted_by","updates","updated_by","see_also","errata_url"].forEach(e=>{if("status"==e){a[e]=a[e].toLowerCase();var t=a[e].split(" "),o=t.length,w="",p=1;for(let e=0;e<o;e++)p<o?w=w+r(t[e])+" ":w+=r(t[e]),p++;a[e]=w}else if("obsoletes"==e||"obsoleted_by"==e||"updates"==e||"updated_by"==e){var g,m="",b=1;g=a[e].length;for(let t=0;t<g;t++)a[e][t]&&(a[e][t]=String(a[e][t]).toLowerCase(),m=b<g?m+"<a href='"+s+"/rfc/".concat(a[e][t])+"'>"+a[e][t].slice(3)+"</a>, ":m+"<a href='"+s+"/rfc/".concat(a[e][t])+"'>"+a[e][t].slice(3)+"</a>",b++);a[e]=m}else if("see_also"==e){var y,L="",C=1;y=a[e].length;for(let t=0;t<y;t++)if(a[e][t]){a[e][t]=String(a[e][t]);var _=a[e][t].slice(0,3),v=a[e][t].slice(3).replace(/^0+/,"");L=C<y?"RFC"!=_?L+"<a href='"+s+"/info/"+_.toLowerCase().concat(v.toLowerCase())+"'>"+_+" "+v+"</a>, ":L+"<a href='"+s+"/info/"+_.toLowerCase().concat(v.toLowerCase())+"'>"+v+"</a>, ":"RFC"!=_?L+"<a href='"+s+"/info/"+_.toLowerCase().concat(v.toLowerCase())+"'>"+_+" "+v+"</a>":L+"<a href='"+s+"/info/"+_.toLowerCase().concat(v.toLowerCase())+"'>"+v+"</a>",C++}a[e]=L}else if("errata_url"==e){var R="";R=a[e]?R+"<a href='"+a[e]+"'>Errata exist</a> | <a href='"+d+"/"+l+"'>Datatracker</a>| <a href='"+n+"/?"+i+"="+f+"&submit="+i+"'>IPR</a> | <a href='"+c+"/"+l+"'>Info page</a>":"<a href='"+d+"/"+l+"'>Datatracker</a> | <a href='"+n+"/?"+i+"="+f+"&submit="+i+"'>IPR</a> | <a href='"+c+"/"+l+"'>Info page</a>",a[e]=R}""!=a[e]?"Errata"==u[e]?h+=`<dt>More info:</dt><dd>${a[e]}</dd>`:h+=`<dt>${u[e]}:</dt><dd>${a[e]}</dd>`:"Errata"==u[e]&&(h+=`<dt>More info:</dt><dd>${a[e]}</dd>`)}),h+="</dl>",e.innerHTML=h}catch(e){console.log(e)}else console.log("Could not locate metadata <div> element");function r(e){return e.charAt(0).toUpperCase()+e.slice(1)}}window.removeEventListener("load",addMetadata),window.addEventListener("load",addMetadata);</script>
</head>
<body class="xml2rfc">
<table class="ears">
<thead><tr>
<td class="left">Internet-Draft</td>
<td class="center">Agent Behavioral Verification</td>
<td class="right">March 2026</td>
</tr></thead>
<tfoot><tr>
<td class="left">Nennemann</td>
<td class="center">Expires 7 September 2026</td>
<td class="right">[Page]</td>
</tr></tfoot>
</table>
<div id="external-metadata" class="document-information"></div>
<div id="internal-metadata" class="document-information">
<dl id="identifiers">
<dt class="label-workgroup">Workgroup:</dt>
<dd class="workgroup">NMOP</dd>
<dt class="label-internet-draft">Internet-Draft:</dt>
<dd class="internet-draft">draft-nennemann-agent-behavioral-verification-00</dd>
<dt class="label-published">Published:</dt>
<dd class="published">
<time datetime="2026-03-06" class="published">6 March 2026</time>
</dd>
<dt class="label-intended-status">Intended Status:</dt>
<dd class="intended-status">Standards Track</dd>
<dt class="label-expires">Expires:</dt>
<dd class="expires"><time datetime="2026-09-07">7 September 2026</time></dd>
<dt class="label-authors">Author:</dt>
<dd class="authors">
<div class="author">
<div class="author-name">C. Nennemann</div>
<div class="org">Independent Researcher</div>
</div>
</dd>
</dl>
</div>
<h1 id="title">Agent Behavioral Verification and Performance Benchmarking</h1>
<section id="section-abstract">
<h2 id="abstract"><a href="#abstract" class="selfRef">Abstract</a></h2>
<p id="section-abstract-1">This document defines protocols for runtime
verification that deployed AI agents behave
according to their declared policies. It also
specifies standardized metrics and a framework
for benchmarking agent performance across
implementations. Behavioral Evidence Tokens
(BETs) extend the Execution Context Token
architecture to provide cryptographically
verifiable proof of policy compliance.
Performance profiles enable objective comparison
of agent capabilities.<a href="#section-abstract-1" class="pilcrow"></a></p>
</section>
<div id="status-of-memo">
<section id="section-boilerplate.1">
<h2 id="name-status-of-this-memo">
<a href="#name-status-of-this-memo" class="section-name selfRef">Status of This Memo</a>
</h2>
<p id="section-boilerplate.1-1">
This Internet-Draft is submitted in full conformance with the
provisions of BCP 78 and BCP 79.<a href="#section-boilerplate.1-1" class="pilcrow"></a></p>
<p id="section-boilerplate.1-2">
Internet-Drafts are working documents of the Internet Engineering Task
Force (IETF). Note that other groups may also distribute working
documents as Internet-Drafts. The list of current Internet-Drafts is
at <span><a href="https://datatracker.ietf.org/drafts/current/">https://datatracker.ietf.org/drafts/current/</a></span>.<a href="#section-boilerplate.1-2" class="pilcrow"></a></p>
<p id="section-boilerplate.1-3">
Internet-Drafts are draft documents valid for a maximum of six months
and may be updated, replaced, or obsoleted by other documents at any
time. It is inappropriate to use Internet-Drafts as reference
material or to cite them other than as "work in progress."<a href="#section-boilerplate.1-3" class="pilcrow"></a></p>
<p id="section-boilerplate.1-4">
This Internet-Draft will expire on 7 September 2026.<a href="#section-boilerplate.1-4" class="pilcrow"></a></p>
</section>
</div>
<div id="copyright">
<section id="section-boilerplate.2">
<h2 id="name-copyright-notice">
<a href="#name-copyright-notice" class="section-name selfRef">Copyright Notice</a>
</h2>
<p id="section-boilerplate.2-1">
Copyright (c) 2026 IETF Trust and the persons identified as the
document authors. All rights reserved.<a href="#section-boilerplate.2-1" class="pilcrow"></a></p>
<p id="section-boilerplate.2-2">
This document is subject to BCP 78 and the IETF Trust's Legal
Provisions Relating to IETF Documents
(<span><a href="https://trustee.ietf.org/license-info">https://trustee.ietf.org/license-info</a></span>) in effect on the date of
publication of this document. Please review these documents
carefully, as they describe your rights and restrictions with
respect to this document. Code Components extracted from this
document must include Revised BSD License text as described in
Section 4.e of the Trust Legal Provisions and are provided without
warranty as described in the Revised BSD License.<a href="#section-boilerplate.2-2" class="pilcrow"></a></p>
</section>
</div>
<div id="toc">
<section id="section-toc.1">
<a href="#" onclick="scroll(0,0)" class="toplink"></a><h2 id="name-table-of-contents">
<a href="#name-table-of-contents" class="section-name selfRef">Table of Contents</a>
</h2>
<nav class="toc"><ul class="compact toc ulBare ulEmpty">
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.1">
<p id="section-toc.1-1.1.1" class="keepWithNext"><a href="#section-1" class="auto internal xref">1</a>.  <a href="#name-introduction" class="internal xref">Introduction</a></p>
</li>
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.2">
<p id="section-toc.1-1.2.1" class="keepWithNext"><a href="#section-2" class="auto internal xref">2</a>.  <a href="#name-terminology" class="internal xref">Terminology</a></p>
</li>
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.3">
<p id="section-toc.1-1.3.1"><a href="#section-3" class="auto internal xref">3</a>.  <a href="#name-behavioral-verification-arc" class="internal xref">Behavioral Verification Architecture</a></p>
<ul class="compact toc ulBare ulEmpty">
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.3.2.1">
<p id="section-toc.1-1.3.2.1.1" class="keepWithNext"><a href="#section-3.1" class="auto internal xref">3.1</a>.  <a href="#name-verification-model-overview" class="internal xref">Verification Model Overview</a></p>
</li>
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.3.2.2">
<p id="section-toc.1-1.3.2.2.1"><a href="#section-3.2" class="auto internal xref">3.2</a>.  <a href="#name-policy-behavior-binding" class="internal xref">Policy-Behavior Binding</a></p>
</li>
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.3.2.3">
<p id="section-toc.1-1.3.2.3.1"><a href="#section-3.3" class="auto internal xref">3.3</a>.  <a href="#name-behavioral-evidence-tokens-" class="internal xref">Behavioral Evidence Tokens (BET)</a></p>
<ul class="compact toc ulBare ulEmpty">
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.3.2.3.2.1">
<p id="section-toc.1-1.3.2.3.2.1.1"><a href="#section-3.3.1" class="auto internal xref">3.3.1</a>.  <a href="#name-bet-lifecycle" class="internal xref">BET Lifecycle</a></p>
</li>
</ul>
</li>
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.3.2.4">
<p id="section-toc.1-1.3.2.4.1"><a href="#section-3.4" class="auto internal xref">3.4</a>.  <a href="#name-runtime-monitoring-protocol" class="internal xref">Runtime Monitoring Protocol</a></p>
<ul class="compact toc ulBare ulEmpty">
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.3.2.4.2.1">
<p id="section-toc.1-1.3.2.4.2.1.1"><a href="#section-3.4.1" class="auto internal xref">3.4.1</a>.  <a href="#name-monitor-placement" class="internal xref">Monitor Placement</a></p>
</li>
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.3.2.4.2.2">
<p id="section-toc.1-1.3.2.4.2.2.1"><a href="#section-3.4.2" class="auto internal xref">3.4.2</a>.  <a href="#name-observation-collection" class="internal xref">Observation Collection</a></p>
</li>
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.3.2.4.2.3">
<p id="section-toc.1-1.3.2.4.2.3.1"><a href="#section-3.4.3" class="auto internal xref">3.4.3</a>.  <a href="#name-evidence-assembly" class="internal xref">Evidence Assembly</a></p>
</li>
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.3.2.4.2.4">
<p id="section-toc.1-1.3.2.4.2.4.1"><a href="#section-3.4.4" class="auto internal xref">3.4.4</a>.  <a href="#name-anomaly-detection-signaling" class="internal xref">Anomaly Detection Signaling</a></p>
</li>
</ul>
</li>
</ul>
</li>
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.4">
<p id="section-toc.1-1.4.1"><a href="#section-4" class="auto internal xref">4</a>.  <a href="#name-performance-benchmarking-fr" class="internal xref">Performance Benchmarking Framework</a></p>
<ul class="compact toc ulBare ulEmpty">
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.4.2.1">
<p id="section-toc.1-1.4.2.1.1"><a href="#section-4.1" class="auto internal xref">4.1</a>.  <a href="#name-standard-metrics" class="internal xref">Standard Metrics</a></p>
</li>
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.4.2.2">
<p id="section-toc.1-1.4.2.2.1"><a href="#section-4.2" class="auto internal xref">4.2</a>.  <a href="#name-benchmark-profiles" class="internal xref">Benchmark Profiles</a></p>
</li>
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.4.2.3">
<p id="section-toc.1-1.4.2.3.1"><a href="#section-4.3" class="auto internal xref">4.3</a>.  <a href="#name-benchmark-execution-protoco" class="internal xref">Benchmark Execution Protocol</a></p>
<ul class="compact toc ulBare ulEmpty">
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.4.2.3.2.1">
<p id="section-toc.1-1.4.2.3.2.1.1"><a href="#section-4.3.1" class="auto internal xref">4.3.1</a>.  <a href="#name-test-harness-requirements" class="internal xref">Test Harness Requirements</a></p>
</li>
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.4.2.3.2.2">
<p id="section-toc.1-1.4.2.3.2.2.1"><a href="#section-4.3.2" class="auto internal xref">4.3.2</a>.  <a href="#name-result-reporting-format" class="internal xref">Result Reporting Format</a></p>
</li>
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.4.2.3.2.3">
<p id="section-toc.1-1.4.2.3.2.3.1"><a href="#section-4.3.3" class="auto internal xref">4.3.3</a>.  <a href="#name-anti-gaming-provisions" class="internal xref">Anti-Gaming Provisions</a></p>
</li>
</ul>
</li>
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.4.2.4">
<p id="section-toc.1-1.4.2.4.1"><a href="#section-4.4" class="auto internal xref">4.4</a>.  <a href="#name-performance-claims-in-ect" class="internal xref">Performance Claims in ECT</a></p>
</li>
</ul>
</li>
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.5">
<p id="section-toc.1-1.5.1"><a href="#section-5" class="auto internal xref">5</a>.  <a href="#name-integration-with-ect" class="internal xref">Integration with ECT</a></p>
</li>
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.6">
<p id="section-toc.1-1.6.1"><a href="#section-6" class="auto internal xref">6</a>.  <a href="#name-security-considerations" class="internal xref">Security Considerations</a></p>
<ul class="compact toc ulBare ulEmpty">
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.6.2.1">
<p id="section-toc.1-1.6.2.1.1"><a href="#section-6.1" class="auto internal xref">6.1</a>.  <a href="#name-adversarial-behavior" class="internal xref">Adversarial Behavior</a></p>
</li>
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.6.2.2">
<p id="section-toc.1-1.6.2.2.1"><a href="#section-6.2" class="auto internal xref">6.2</a>.  <a href="#name-monitor-compromise" class="internal xref">Monitor Compromise</a></p>
</li>
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.6.2.3">
<p id="section-toc.1-1.6.2.3.1"><a href="#section-6.3" class="auto internal xref">6.3</a>.  <a href="#name-benchmark-manipulation" class="internal xref">Benchmark Manipulation</a></p>
</li>
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.6.2.4">
<p id="section-toc.1-1.6.2.4.1"><a href="#section-6.4" class="auto internal xref">6.4</a>.  <a href="#name-privacy-of-behavioral-evide" class="internal xref">Privacy of Behavioral Evidence</a></p>
</li>
</ul>
</li>
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.7">
<p id="section-toc.1-1.7.1"><a href="#section-7" class="auto internal xref">7</a>.  <a href="#name-iana-considerations" class="internal xref">IANA Considerations</a></p>
<ul class="compact toc ulBare ulEmpty">
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.7.2.1">
<p id="section-toc.1-1.7.2.1.1"><a href="#section-7.1" class="auto internal xref">7.1</a>.  <a href="#name-ect-extension-claim-keys" class="internal xref">ECT Extension Claim Keys</a></p>
</li>
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.7.2.2">
<p id="section-toc.1-1.7.2.2.1"><a href="#section-7.2" class="auto internal xref">7.2</a>.  <a href="#name-benchmark-profile-media-typ" class="internal xref">Benchmark Profile Media Type</a></p>
</li>
</ul>
</li>
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.8">
<p id="section-toc.1-1.8.1"><a href="#section-8" class="auto internal xref">8</a>.  <a href="#name-references" class="internal xref">References</a></p>
<ul class="compact toc ulBare ulEmpty">
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.8.2.1">
<p id="section-toc.1-1.8.2.1.1"><a href="#section-8.1" class="auto internal xref">8.1</a>.  <a href="#name-normative-references" class="internal xref">Normative References</a></p>
</li>
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.8.2.2">
<p id="section-toc.1-1.8.2.2.1"><a href="#section-8.2" class="auto internal xref">8.2</a>.  <a href="#name-informative-references" class="internal xref">Informative References</a></p>
</li>
</ul>
</li>
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.9">
<p id="section-toc.1-1.9.1"><a href="#appendix-A" class="auto internal xref"></a><a href="#name-acknowledgments" class="internal xref">Acknowledgments</a></p>
</li>
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.10">
<p id="section-toc.1-1.10.1"><a href="#appendix-B" class="auto internal xref"></a><a href="#name-authors-address" class="internal xref">Author's Address</a></p>
</li>
</ul>
</nav>
</section>
</div>
<div id="introduction">
<section id="section-1">
<h2 id="name-introduction">
<a href="#section-1" class="section-number selfRef">1. </a><a href="#name-introduction" class="section-name selfRef">Introduction</a>
</h2>
<p id="section-1-1">Autonomous AI agents increasingly operate in
networked environments where they make decisions,
invoke tools, and delegate tasks to other agents.
Operators and relying parties need assurance that
these agents behave according to their declared
policies at runtime, not merely at deployment
time.<a href="#section-1-1" class="pilcrow"></a></p>
<p id="section-1-2"><span>[<a href="#I-D.nennemann-agent-gap-analysis" class="cite xref">I-D.nennemann-agent-gap-analysis</a>]</span> identifies
two critical gaps in the current standards
landscape:<a href="#section-1-2" class="pilcrow"></a></p>
<ul class="normal">
<li class="normal" id="section-1-3.1">
<p id="section-1-3.1.1">Gap 1 (Behavioral Verification): Agents
declare policies in their Execution Context
Tokens but no standardized mechanism exists to
verify that runtime behavior matches those
declarations.<a href="#section-1-3.1.1" class="pilcrow"></a></p>
</li>
<li class="normal" id="section-1-3.2">
<p id="section-1-3.2.1">Gap 11 (Performance Benchmarking): No
standardized way exists to compare agent
implementations objectively across dimensions
such as task completion, latency, accuracy,
and safety compliance.<a href="#section-1-3.2.1" class="pilcrow"></a></p>
</li>
</ul>
<p id="section-1-4">This document addresses both gaps by defining:<a href="#section-1-4" class="pilcrow"></a></p>
<ol start="1" type="1" class="normal type-1" id="section-1-5">
<li id="section-1-5.1">
<p id="section-1-5.1.1">A behavioral verification architecture
aligned with the Remote Attestation Procedures
(RATS) framework <span>[<a href="#RFC9334" class="cite xref">RFC9334</a>]</span>.<a href="#section-1-5.1.1" class="pilcrow"></a></p>
</li>
<li id="section-1-5.2">
<p id="section-1-5.2.1">Behavioral Evidence Tokens (BETs) that extend
the Execution Context Token (ECT)
<span>[<a href="#I-D.nennemann-wimse-ect" class="cite xref">I-D.nennemann-wimse-ect</a>]</span> with runtime
compliance claims.<a href="#section-1-5.2.1" class="pilcrow"></a></p>
</li>
<li id="section-1-5.3">
<p id="section-1-5.3.1">A performance benchmarking framework with
standard metrics, benchmark profiles, and an
execution protocol.<a href="#section-1-5.3.1" class="pilcrow"></a></p>
</li>
</ol>
</section>
</div>
<div id="terminology">
<section id="section-2">
<h2 id="name-terminology">
<a href="#section-2" class="section-number selfRef">2. </a><a href="#name-terminology" class="section-name selfRef">Terminology</a>
</h2>
<p id="section-2-1">The key words "<span class="bcp14">MUST</span>", "<span class="bcp14">MUST NOT</span>", "<span class="bcp14">REQUIRED</span>", "<span class="bcp14">SHALL</span>", "<span class="bcp14">SHALL NOT</span>", "<span class="bcp14">SHOULD</span>", "<span class="bcp14">SHOULD NOT</span>", "<span class="bcp14">RECOMMENDED</span>", "<span class="bcp14">NOT RECOMMENDED</span>",
"<span class="bcp14">MAY</span>", and "<span class="bcp14">OPTIONAL</span>" in this document are to be interpreted as
described in BCP 14 <span>[<a href="#RFC2119" class="cite xref">RFC2119</a>]</span> <span>[<a href="#RFC8174" class="cite xref">RFC8174</a>]</span> when, and only when, they
appear in all capitals, as shown here.<a href="#section-2-1" class="pilcrow"></a></p>
<p id="section-2-2">The following terms are used in this document:<a href="#section-2-2" class="pilcrow"></a></p>
<span class="break"></span><dl class="dlParallel" id="section-2-3">
<dt id="section-2-3.1">Behavioral Attestation:</dt>
<dd style="margin-left: 1.5em" id="section-2-3.2">
<p id="section-2-3.2.1">The process of generating verifiable evidence
that an agent's runtime actions conform to its
declared policies.<a href="#section-2-3.2.1" class="pilcrow"></a></p>
</dd>
<dd class="break"></dd>
<dt id="section-2-3.3">Policy-Behavior Binding:</dt>
<dd style="margin-left: 1.5em" id="section-2-3.4">
<p id="section-2-3.4.1">A formal linkage between a declared policy in
an agent's ECT and observable runtime actions
that demonstrate compliance with that policy.<a href="#section-2-3.4.1" class="pilcrow"></a></p>
</dd>
<dd class="break"></dd>
<dt id="section-2-3.5">Behavioral Evidence Token (BET):</dt>
<dd style="margin-left: 1.5em" id="section-2-3.6">
<p id="section-2-3.6.1">A signed token containing claims about an
agent's observed runtime behavior relative to
its declared policies. BETs extend the ECT
architecture.<a href="#section-2-3.6.1" class="pilcrow"></a></p>
</dd>
<dd class="break"></dd>
<dt id="section-2-3.7">Runtime Monitor:</dt>
<dd style="margin-left: 1.5em" id="section-2-3.8">
<p id="section-2-3.8.1">A component that observes agent actions and
collects evidence for behavioral attestation.<a href="#section-2-3.8.1" class="pilcrow"></a></p>
</dd>
<dd class="break"></dd>
<dt id="section-2-3.9">Benchmark Suite:</dt>
<dd style="margin-left: 1.5em" id="section-2-3.10">
<p id="section-2-3.10.1">A collection of standardized test scenarios
designed to evaluate agent performance across
defined metrics.<a href="#section-2-3.10.1" class="pilcrow"></a></p>
</dd>
<dd class="break"></dd>
<dt id="section-2-3.11">Performance Profile:</dt>
<dd style="margin-left: 1.5em" id="section-2-3.12">
<p id="section-2-3.12.1">A structured record of benchmark results for
a specific agent implementation.<a href="#section-2-3.12.1" class="pilcrow"></a></p>
</dd>
<dd class="break"></dd>
</dl>
</section>
</div>
<div id="behavioral-verification-architecture">
<section id="section-3">
<h2 id="name-behavioral-verification-arc">
<a href="#section-3" class="section-number selfRef">3. </a><a href="#name-behavioral-verification-arc" class="section-name selfRef">Behavioral Verification Architecture</a>
</h2>
<div id="verification-model-overview">
<section id="section-3.1">
<h3 id="name-verification-model-overview">
<a href="#section-3.1" class="section-number selfRef">3.1. </a><a href="#name-verification-model-overview" class="section-name selfRef">Verification Model Overview</a>
</h3>
<p id="section-3.1-1">The behavioral verification architecture aligns
with the RATS <span>[<a href="#RFC9334" class="cite xref">RFC9334</a>]</span> roles of Attester,
Verifier, and Relying Party. A Runtime Monitor
collects evidence of agent actions and produces
Behavioral Evidence Tokens.<a href="#section-3.1-1" class="pilcrow"></a></p>
<span id="name-behavioral-verification-arch"></span><div id="fig-arch">
<figure id="figure-1">
<div class="alignLeft art-text artwork" id="section-3.1-2.1">
<pre>
+-------------+ +---------+
| Agent |------&gt;| Runtime |
| (Attester) |actions| Monitor |
+-------------+ +----+----+
|
evidence
|
+----v----+
| BET |
| Creator |
+----+----+
|
BET
|
+---------v---------+
| Verifier |
| (Policy Engine) |
+---------+---------+
|
attestation result
|
+---------v---------+
| Relying Party |
| (Orchestrator / |
| Operator) |
+-------------------+
</pre>
</div>
<figcaption><a href="#figure-1" class="selfRef">Figure 1</a>:
<a href="#name-behavioral-verification-arch" class="selfRef">Behavioral Verification Architecture</a>
</figcaption></figure>
</div>
<p id="section-3.1-3">The architecture supports two modes of
operation:<a href="#section-3.1-3" class="pilcrow"></a></p>
<ul class="normal">
<li class="normal" id="section-3.1-4.1">
<p id="section-3.1-4.1.1">Continuous Monitoring: The Runtime Monitor
observes all agent actions in real time and
generates BETs at configurable intervals or
upon policy-relevant events.<a href="#section-3.1-4.1.1" class="pilcrow"></a></p>
</li>
<li class="normal" id="section-3.1-4.2">
<p id="section-3.1-4.2.1">Point-in-Time Attestation: A Verifier
requests behavioral evidence for a specific
time window, and the Monitor assembles a BET
covering that period.<a href="#section-3.1-4.2.1" class="pilcrow"></a></p>
</li>
</ul>
</section>
</div>
<div id="policy-behavior-binding">
<section id="section-3.2">
<h3 id="name-policy-behavior-binding">
<a href="#section-3.2" class="section-number selfRef">3.2. </a><a href="#name-policy-behavior-binding" class="section-name selfRef">Policy-Behavior Binding</a>
</h3>
<p id="section-3.2-1">A Policy-Behavior Binding declares the expected
behaviors associated with a policy and the
observable actions that constitute compliance.<a href="#section-3.2-1" class="pilcrow"></a></p>
<p id="section-3.2-2">The binding is expressed as a JSON object:<a href="#section-3.2-2" class="pilcrow"></a></p>
<span id="name-policy-behavior-binding-str"></span><div id="fig-binding">
<figure id="figure-2">
<div class="lang-json sourcecode" id="section-3.2-3.1">
<pre>
{
"policy_id": "urn:example:policy:data-access",
"version": "1.0",
"expected_behaviors": [
{
"behavior_id": "bhv-001",
"description": "Agent accesses only
authorized data sources",
"observable_actions": [
"data_source_access"
],
"compliance_criteria": {
"type": "allowlist",
"values": [
"urn:example:ds:approved-1",
"urn:example:ds:approved-2"
]
}
}
],
"evaluation_mode": "continuous"
}
</pre>
</div>
<figcaption><a href="#figure-2" class="selfRef">Figure 2</a>:
<a href="#name-policy-behavior-binding-str" class="selfRef">Policy-Behavior Binding Structure</a>
</figcaption></figure>
</div>
<p id="section-3.2-4">Each binding <span class="bcp14">MUST</span> include:<a href="#section-3.2-4" class="pilcrow"></a></p>
<ul class="normal">
<li class="normal" id="section-3.2-5.1">
<p id="section-3.2-5.1.1"><code>policy_id</code>: A URI identifying the policy.<a href="#section-3.2-5.1.1" class="pilcrow"></a></p>
</li>
<li class="normal" id="section-3.2-5.2">
<p id="section-3.2-5.2.1"><code>expected_behaviors</code>: An array of behavior
descriptors.<a href="#section-3.2-5.2.1" class="pilcrow"></a></p>
</li>
<li class="normal" id="section-3.2-5.3">
<p id="section-3.2-5.3.1"><code>evaluation_mode</code>: Either "continuous" or
"on_demand".<a href="#section-3.2-5.3.1" class="pilcrow"></a></p>
</li>
</ul>
<p id="section-3.2-6">Each behavior descriptor <span class="bcp14">MUST</span> include:<a href="#section-3.2-6" class="pilcrow"></a></p>
<ul class="normal">
<li class="normal" id="section-3.2-7.1">
<p id="section-3.2-7.1.1"><code>behavior_id</code>: A unique identifier.<a href="#section-3.2-7.1.1" class="pilcrow"></a></p>
</li>
<li class="normal" id="section-3.2-7.2">
<p id="section-3.2-7.2.1"><code>observable_actions</code>: Action types the monitor
<span class="bcp14">MUST</span> observe.<a href="#section-3.2-7.2.1" class="pilcrow"></a></p>
</li>
<li class="normal" id="section-3.2-7.3">
<p id="section-3.2-7.3.1"><code>compliance_criteria</code>: The conditions under
which the behavior is considered compliant.<a href="#section-3.2-7.3.1" class="pilcrow"></a></p>
</li>
</ul>
</section>
</div>
<div id="behavioral-evidence-tokens-bet">
<section id="section-3.3">
<h3 id="name-behavioral-evidence-tokens-">
<a href="#section-3.3" class="section-number selfRef">3.3. </a><a href="#name-behavioral-evidence-tokens-" class="section-name selfRef">Behavioral Evidence Tokens (BET)</a>
</h3>
<p id="section-3.3-1">A Behavioral Evidence Token is a JSON Web Token
(JWT) <span>[<a href="#RFC7519" class="cite xref">RFC7519</a>]</span> signed using JSON Web Signature
(JWS) <span>[<a href="#RFC7515" class="cite xref">RFC7515</a>]</span>. BETs extend the ECT claim
set with behavioral verification claims.<a href="#section-3.3-1" class="pilcrow"></a></p>
<p id="section-3.3-2">The following new claims are defined:<a href="#section-3.3-2" class="pilcrow"></a></p>
<span class="break"></span><dl class="dlParallel" id="section-3.3-3">
<dt id="section-3.3-3.1">
<code>bhv_policy</code>:</dt>
<dd style="margin-left: 1.5em" id="section-3.3-3.2">
<p id="section-3.3-3.2.1"><span class="bcp14">REQUIRED</span>. A URI reference to the policy being
verified.<a href="#section-3.3-3.2.1" class="pilcrow"></a></p>
</dd>
<dd class="break"></dd>
<dt id="section-3.3-3.3">
<code>bhv_result</code>:</dt>
<dd style="margin-left: 1.5em" id="section-3.3-3.4">
<p id="section-3.3-3.4.1"><span class="bcp14">REQUIRED</span>. The verification result. One of
"pass", "fail", or "partial".<a href="#section-3.3-3.4.1" class="pilcrow"></a></p>
</dd>
<dd class="break"></dd>
<dt id="section-3.3-3.5">
<code>bhv_evidence</code>:</dt>
<dd style="margin-left: 1.5em" id="section-3.3-3.6">
<p id="section-3.3-3.6.1"><span class="bcp14">REQUIRED</span>. A base64url-encoded hash (SHA-256)
of the collected observable actions during the
observation window.<a href="#section-3.3-3.6.1" class="pilcrow"></a></p>
</dd>
<dd class="break"></dd>
<dt id="section-3.3-3.7">
<code>bhv_window</code>:</dt>
<dd style="margin-left: 1.5em" id="section-3.3-3.8">
<p id="section-3.3-3.8.1"><span class="bcp14">REQUIRED</span>. A JSON object with <code>start</code> and
<code>end</code> fields containing NumericDate values
(as defined in <span>[<a href="#RFC7519" class="cite xref">RFC7519</a>]</span>) representing the
observation period.<a href="#section-3.3-3.8.1" class="pilcrow"></a></p>
</dd>
<dd class="break"></dd>
<dt id="section-3.3-3.9">
<code>bhv_details</code>:</dt>
<dd style="margin-left: 1.5em" id="section-3.3-3.10">
<p id="section-3.3-3.10.1"><span class="bcp14">OPTIONAL</span>. An array of per-behavior results
with <code>behavior_id</code> and individual <code>result</code>
values.<a href="#section-3.3-3.10.1" class="pilcrow"></a></p>
</dd>
<dd class="break"></dd>
</dl>
<p id="section-3.3-4">Example BET payload:<a href="#section-3.3-4" class="pilcrow"></a></p>
<span id="name-example-bet-payload"></span><div id="fig-bet">
<figure id="figure-3">
<div class="lang-json sourcecode" id="section-3.3-5.1">
<pre>
{
"iss": "urn:example:monitor:m-001",
"sub": "urn:example:agent:agent-42",
"iat": 1700000000,
"exp": 1700003600,
"bhv_policy": "urn:example:policy:data-access",
"bhv_result": "pass",
"bhv_evidence": "dGhpcyBpcyBhIGhhc2g...",
"bhv_window": {
"start": 1699996400,
"end": 1700000000
},
"bhv_details": [
{
"behavior_id": "bhv-001",
"result": "pass"
}
]
}
</pre>
</div>
<figcaption><a href="#figure-3" class="selfRef">Figure 3</a>:
<a href="#name-example-bet-payload" class="selfRef">Example BET Payload</a>
</figcaption></figure>
</div>
<div id="bet-lifecycle">
<section id="section-3.3.1">
<h4 id="name-bet-lifecycle">
<a href="#section-3.3.1" class="section-number selfRef">3.3.1. </a><a href="#name-bet-lifecycle" class="section-name selfRef">BET Lifecycle</a>
</h4>
<p id="section-3.3.1-1">The lifecycle of a Behavioral Evidence Token
consists of three phases:<a href="#section-3.3.1-1" class="pilcrow"></a></p>
<ol start="1" type="1" class="normal type-1" id="section-3.3.1-2">
<li id="section-3.3.1-2.1">
<p id="section-3.3.1-2.1.1">Creation: The Runtime Monitor collects
evidence of agent actions, evaluates them
against the Policy-Behavior Binding, and
constructs a BET with the appropriate claims.
The BET is signed by the Monitor's key.<a href="#section-3.3.1-2.1.1" class="pilcrow"></a></p>
</li>
<li id="section-3.3.1-2.2">
<p id="section-3.3.1-2.2.1">Submission: The signed BET is submitted to
the Verifier. Submission <span class="bcp14">MAY</span> occur via a
push model (Monitor sends to Verifier) or a
pull model (Verifier requests from Monitor).<a href="#section-3.3.1-2.2.1" class="pilcrow"></a></p>
</li>
<li id="section-3.3.1-2.3">
<p id="section-3.3.1-2.3.1">Verification: The Verifier validates the BET
signature, checks the claims against its
reference policies, and produces an
attestation result for the Relying Party.<a href="#section-3.3.1-2.3.1" class="pilcrow"></a></p>
</li>
</ol>
</section>
</div>
</section>
</div>
<div id="runtime-monitoring-protocol">
<section id="section-3.4">
<h3 id="name-runtime-monitoring-protocol">
<a href="#section-3.4" class="section-number selfRef">3.4. </a><a href="#name-runtime-monitoring-protocol" class="section-name selfRef">Runtime Monitoring Protocol</a>
</h3>
<div id="monitor-placement">
<section id="section-3.4.1">
<h4 id="name-monitor-placement">
<a href="#section-3.4.1" class="section-number selfRef">3.4.1. </a><a href="#name-monitor-placement" class="section-name selfRef">Monitor Placement</a>
</h4>
<p id="section-3.4.1-1">Runtime Monitors <span class="bcp14">MAY</span> be deployed in one of three
configurations:<a href="#section-3.4.1-1" class="pilcrow"></a></p>
<span class="break"></span><dl class="dlParallel" id="section-3.4.1-2">
<dt id="section-3.4.1-2.1">Inline:</dt>
<dd style="margin-left: 1.5em" id="section-3.4.1-2.2">
<p id="section-3.4.1-2.2.1">The Monitor intercepts all agent
communications as a proxy. This provides
complete visibility but adds latency.<a href="#section-3.4.1-2.2.1" class="pilcrow"></a></p>
</dd>
<dd class="break"></dd>
<dt id="section-3.4.1-2.3">Sidecar:</dt>
<dd style="margin-left: 1.5em" id="section-3.4.1-2.4">
<p id="section-3.4.1-2.4.1">The Monitor runs alongside the agent process
and receives copies of all actions via a local
interface. This minimizes latency while
maintaining visibility.<a href="#section-3.4.1-2.4.1" class="pilcrow"></a></p>
</dd>
<dd class="break"></dd>
<dt id="section-3.4.1-2.5">External:</dt>
<dd style="margin-left: 1.5em" id="section-3.4.1-2.6">
<p id="section-3.4.1-2.6.1">The Monitor operates as a separate service
that receives action logs asynchronously.
This provides the least overhead but may miss
real-time events.<a href="#section-3.4.1-2.6.1" class="pilcrow"></a></p>
</dd>
<dd class="break"></dd>
</dl>
</section>
</div>
<div id="observation-collection">
<section id="section-3.4.2">
<h4 id="name-observation-collection">
<a href="#section-3.4.2" class="section-number selfRef">3.4.2. </a><a href="#name-observation-collection" class="section-name selfRef">Observation Collection</a>
</h4>
<p id="section-3.4.2-1">The Monitor <span class="bcp14">MUST</span> maintain a time-ordered log of
observed actions. Each log entry <span class="bcp14">MUST</span> contain:<a href="#section-3.4.2-1" class="pilcrow"></a></p>
<ul class="normal">
<li class="normal" id="section-3.4.2-2.1">
<p id="section-3.4.2-2.1.1">Timestamp (NumericDate)<a href="#section-3.4.2-2.1.1" class="pilcrow"></a></p>
</li>
<li class="normal" id="section-3.4.2-2.2">
<p id="section-3.4.2-2.2.1">Action type<a href="#section-3.4.2-2.2.1" class="pilcrow"></a></p>
</li>
<li class="normal" id="section-3.4.2-2.3">
<p id="section-3.4.2-2.3.1">Action target (URI)<a href="#section-3.4.2-2.3.1" class="pilcrow"></a></p>
</li>
<li class="normal" id="section-3.4.2-2.4">
<p id="section-3.4.2-2.4.1">Action parameters (opaque to the Monitor)<a href="#section-3.4.2-2.4.1" class="pilcrow"></a></p>
</li>
<li class="normal" id="section-3.4.2-2.5">
<p id="section-3.4.2-2.5.1">Agent identifier<a href="#section-3.4.2-2.5.1" class="pilcrow"></a></p>
</li>
</ul>
</section>
</div>
<div id="evidence-assembly">
<section id="section-3.4.3">
<h4 id="name-evidence-assembly">
<a href="#section-3.4.3" class="section-number selfRef">3.4.3. </a><a href="#name-evidence-assembly" class="section-name selfRef">Evidence Assembly</a>
</h4>
<p id="section-3.4.3-1">When assembling evidence for a BET, the Monitor
<span class="bcp14">MUST</span>:<a href="#section-3.4.3-1" class="pilcrow"></a></p>
<ol start="1" type="1" class="normal type-1" id="section-3.4.3-2">
<li id="section-3.4.3-2.1">
<p id="section-3.4.3-2.1.1">Select all log entries within the observation
window.<a href="#section-3.4.3-2.1.1" class="pilcrow"></a></p>
</li>
<li id="section-3.4.3-2.2">
<p id="section-3.4.3-2.2.1">Compute a SHA-256 hash over the canonical
JSON serialization of the selected entries.<a href="#section-3.4.3-2.2.1" class="pilcrow"></a></p>
</li>
<li id="section-3.4.3-2.3">
<p id="section-3.4.3-2.3.1">Evaluate each entry against the applicable
Policy-Behavior Bindings.<a href="#section-3.4.3-2.3.1" class="pilcrow"></a></p>
</li>
<li id="section-3.4.3-2.4">
<p id="section-3.4.3-2.4.1">Determine the aggregate <code>bhv_result</code>.<a href="#section-3.4.3-2.4.1" class="pilcrow"></a></p>
</li>
</ol>
</section>
</div>
<div id="anomaly-detection-signaling">
<section id="section-3.4.4">
<h4 id="name-anomaly-detection-signaling">
<a href="#section-3.4.4" class="section-number selfRef">3.4.4. </a><a href="#name-anomaly-detection-signaling" class="section-name selfRef">Anomaly Detection Signaling</a>
</h4>
<p id="section-3.4.4-1">When the Monitor detects behavior that violates
a Policy-Behavior Binding, it <span class="bcp14">MUST</span>:<a href="#section-3.4.4-1" class="pilcrow"></a></p>
<ol start="1" type="1" class="normal type-1" id="section-3.4.4-2">
<li id="section-3.4.4-2.1">
<p id="section-3.4.4-2.1.1">Generate a BET with <code>bhv_result</code> set to
"fail" or "partial".<a href="#section-3.4.4-2.1.1" class="pilcrow"></a></p>
</li>
<li id="section-3.4.4-2.2">
<p id="section-3.4.4-2.2.1">Signal the anomaly to the Verifier
immediately, regardless of the configured
reporting interval.<a href="#section-3.4.4-2.2.1" class="pilcrow"></a></p>
</li>
<li id="section-3.4.4-2.3">
<p id="section-3.4.4-2.3.1">Optionally signal the agent's orchestrator
to enable corrective action.<a href="#section-3.4.4-2.3.1" class="pilcrow"></a></p>
</li>
</ol>
</section>
</div>
</section>
</div>
</section>
</div>
<div id="performance-benchmarking-framework">
<section id="section-4">
<h2 id="name-performance-benchmarking-fr">
<a href="#section-4" class="section-number selfRef">4. </a><a href="#name-performance-benchmarking-fr" class="section-name selfRef">Performance Benchmarking Framework</a>
</h2>
<div id="standard-metrics">
<section id="section-4.1">
<h3 id="name-standard-metrics">
<a href="#section-4.1" class="section-number selfRef">4.1. </a><a href="#name-standard-metrics" class="section-name selfRef">Standard Metrics</a>
</h3>
<p id="section-4.1-1">The following metrics are defined for agent
performance benchmarking:<a href="#section-4.1-1" class="pilcrow"></a></p>
<span class="break"></span><dl class="dlParallel" id="section-4.1-2">
<dt id="section-4.1-2.1">Task Completion Rate (TCR):</dt>
<dd style="margin-left: 1.5em" id="section-4.1-2.2">
<p id="section-4.1-2.2.1">The ratio of successfully completed tasks to
total tasks attempted. Unit: percentage (%).
Measured over a complete benchmark suite run.<a href="#section-4.1-2.2.1" class="pilcrow"></a></p>
</dd>
<dd class="break"></dd>
<dt id="section-4.1-2.3">Task Latency (TL):</dt>
<dd style="margin-left: 1.5em" id="section-4.1-2.4">
<p id="section-4.1-2.4.1">The time elapsed from task assignment to task
completion. Unit: milliseconds (ms).
Reported as p50, p95, and p99 percentiles.<a href="#section-4.1-2.4.1" class="pilcrow"></a></p>
</dd>
<dd class="break"></dd>
<dt id="section-4.1-2.5">Task Accuracy (TA):</dt>
<dd style="margin-left: 1.5em" id="section-4.1-2.6">
<p id="section-4.1-2.6.1">The degree to which task outputs match
expected results. Unit: percentage (%).
Measured using benchmark-specific evaluation
functions.<a href="#section-4.1-2.6.1" class="pilcrow"></a></p>
</dd>
<dd class="break"></dd>
<dt id="section-4.1-2.7">Resource Efficiency (RE):</dt>
<dd style="margin-left: 1.5em" id="section-4.1-2.8">
<p id="section-4.1-2.8.1">The computational resources consumed per task.
Unit: normalized resource units (NRU).
Includes CPU, memory, and network I/O.<a href="#section-4.1-2.8.1" class="pilcrow"></a></p>
</dd>
<dd class="break"></dd>
<dt id="section-4.1-2.9">Safety Compliance Score (SCS):</dt>
<dd style="margin-left: 1.5em" id="section-4.1-2.10">
<p id="section-4.1-2.10.1">The ratio of tasks completed without safety
policy violations to total tasks.
Unit: percentage (%).<a href="#section-4.1-2.10.1" class="pilcrow"></a></p>
</dd>
<dd class="break"></dd>
<dt id="section-4.1-2.11">Delegation Success Rate (DSR):</dt>
<dd style="margin-left: 1.5em" id="section-4.1-2.12">
<p id="section-4.1-2.12.1">The ratio of successful delegations to total
delegation attempts. Unit: percentage (%).
Applicable only to multi-agent scenarios.<a href="#section-4.1-2.12.1" class="pilcrow"></a></p>
</dd>
<dd class="break"></dd>
</dl>
</section>
</div>
<div id="benchmark-profiles">
<section id="section-4.2">
<h3 id="name-benchmark-profiles">
<a href="#section-4.2" class="section-number selfRef">4.2. </a><a href="#name-benchmark-profiles" class="section-name selfRef">Benchmark Profiles</a>
</h3>
<p id="section-4.2-1">A Benchmark Profile defines a standardized set
of test scenarios for a specific agent category.
Profiles are expressed as JSON objects:<a href="#section-4.2-1" class="pilcrow"></a></p>
<span id="name-benchmark-profile-structure"></span><div id="fig-profile">
<figure id="figure-4">
<div class="lang-json sourcecode" id="section-4.2-2.1">
<pre>
{
"profile_id": "urn:ietf:bench:general-v1",
"profile_name": "General Agent Benchmark",
"version": "1.0",
"agent_category": "general-purpose",
"scenarios": [
{
"scenario_id": "s-001",
"description": "Simple data retrieval",
"difficulty": "basic",
"metrics": ["TCR", "TL", "TA"],
"timeout_ms": 30000,
"expected_output_schema": "..."
}
],
"scoring": {
"weights": {
"TCR": 0.3,
"TL": 0.2,
"TA": 0.3,
"SCS": 0.2
}
}
}
</pre>
</div>
<figcaption><a href="#figure-4" class="selfRef">Figure 4</a>:
<a href="#name-benchmark-profile-structure" class="selfRef">Benchmark Profile Structure</a>
</figcaption></figure>
</div>
<p id="section-4.2-3">Predefined profiles <span class="bcp14">SHOULD</span> be registered for
common agent types including:<a href="#section-4.2-3" class="pilcrow"></a></p>
<ul class="normal">
<li class="normal" id="section-4.2-4.1">
<p id="section-4.2-4.1.1">General-purpose agents<a href="#section-4.2-4.1.1" class="pilcrow"></a></p>
</li>
<li class="normal" id="section-4.2-4.2">
<p id="section-4.2-4.2.1">Code generation agents<a href="#section-4.2-4.2.1" class="pilcrow"></a></p>
</li>
<li class="normal" id="section-4.2-4.3">
<p id="section-4.2-4.3.1">Data analysis agents<a href="#section-4.2-4.3.1" class="pilcrow"></a></p>
</li>
<li class="normal" id="section-4.2-4.4">
<p id="section-4.2-4.4.1">Network management agents<a href="#section-4.2-4.4.1" class="pilcrow"></a></p>
</li>
</ul>
</section>
</div>
<div id="benchmark-execution-protocol">
<section id="section-4.3">
<h3 id="name-benchmark-execution-protoco">
<a href="#section-4.3" class="section-number selfRef">4.3. </a><a href="#name-benchmark-execution-protoco" class="section-name selfRef">Benchmark Execution Protocol</a>
</h3>
<div id="test-harness-requirements">
<section id="section-4.3.1">
<h4 id="name-test-harness-requirements">
<a href="#section-4.3.1" class="section-number selfRef">4.3.1. </a><a href="#name-test-harness-requirements" class="section-name selfRef">Test Harness Requirements</a>
</h4>
<p id="section-4.3.1-1">A conformant test harness <span class="bcp14">MUST</span>:<a href="#section-4.3.1-1" class="pilcrow"></a></p>
<ol start="1" type="1" class="normal type-1" id="section-4.3.1-2">
<li id="section-4.3.1-2.1">
<p id="section-4.3.1-2.1.1">Execute all scenarios in the benchmark
profile in a controlled environment.<a href="#section-4.3.1-2.1.1" class="pilcrow"></a></p>
</li>
<li id="section-4.3.1-2.2">
<p id="section-4.3.1-2.2.1">Isolate agent instances from external
resources not specified in the scenario.<a href="#section-4.3.1-2.2.1" class="pilcrow"></a></p>
</li>
<li id="section-4.3.1-2.3">
<p id="section-4.3.1-2.3.1">Record all metrics defined in the profile.<a href="#section-4.3.1-2.3.1" class="pilcrow"></a></p>
</li>
<li id="section-4.3.1-2.4">
<p id="section-4.3.1-2.4.1">Produce a benchmark result document.<a href="#section-4.3.1-2.4.1" class="pilcrow"></a></p>
</li>
</ol>
</section>
</div>
<div id="result-reporting-format">
<section id="section-4.3.2">
<h4 id="name-result-reporting-format">
<a href="#section-4.3.2" class="section-number selfRef">4.3.2. </a><a href="#name-result-reporting-format" class="section-name selfRef">Result Reporting Format</a>
</h4>
<p id="section-4.3.2-1">Benchmark results <span class="bcp14">MUST</span> be reported as a JSON
object containing:<a href="#section-4.3.2-1" class="pilcrow"></a></p>
<ul class="normal">
<li class="normal" id="section-4.3.2-2.1">
<p id="section-4.3.2-2.1.1"><code>profile_id</code>: The benchmark profile used.<a href="#section-4.3.2-2.1.1" class="pilcrow"></a></p>
</li>
<li class="normal" id="section-4.3.2-2.2">
<p id="section-4.3.2-2.2.1"><code>agent_id</code>: Identifier of the tested agent.<a href="#section-4.3.2-2.2.1" class="pilcrow"></a></p>
</li>
<li class="normal" id="section-4.3.2-2.3">
<p id="section-4.3.2-2.3.1"><code>timestamp</code>: Time of benchmark execution.<a href="#section-4.3.2-2.3.1" class="pilcrow"></a></p>
</li>
<li class="normal" id="section-4.3.2-2.4">
<p id="section-4.3.2-2.4.1"><code>results</code>: Per-scenario metric values.<a href="#section-4.3.2-2.4.1" class="pilcrow"></a></p>
</li>
<li class="normal" id="section-4.3.2-2.5">
<p id="section-4.3.2-2.5.1"><code>aggregate</code>: Weighted aggregate scores.<a href="#section-4.3.2-2.5.1" class="pilcrow"></a></p>
</li>
</ul>
</section>
</div>
<div id="anti-gaming-provisions">
<section id="section-4.3.3">
<h4 id="name-anti-gaming-provisions">
<a href="#section-4.3.3" class="section-number selfRef">4.3.3. </a><a href="#name-anti-gaming-provisions" class="section-name selfRef">Anti-Gaming Provisions</a>
</h4>
<p id="section-4.3.3-1">To prevent agents from gaming benchmark results,
the following provisions apply:<a href="#section-4.3.3-1" class="pilcrow"></a></p>
<ol start="1" type="1" class="normal type-1" id="section-4.3.3-2">
<li id="section-4.3.3-2.1">
<p id="section-4.3.3-2.1.1">Randomized Scenarios: Test harnesses <span class="bcp14">MUST</span>
randomize scenario ordering and <span class="bcp14">MAY</span>
introduce minor variations in scenario
parameters.<a href="#section-4.3.3-2.1.1" class="pilcrow"></a></p>
</li>
<li id="section-4.3.3-2.2">
<p id="section-4.3.3-2.2.1">Blind Evaluation: The agent under test
<span class="bcp14">MUST NOT</span> have access to the expected
outputs or evaluation functions.<a href="#section-4.3.3-2.2.1" class="pilcrow"></a></p>
</li>
<li id="section-4.3.3-2.3">
<p id="section-4.3.3-2.3.1">Holdback Scenarios: Benchmark profiles
<span class="bcp14">SHOULD</span> include scenarios not disclosed to
agent developers.<a href="#section-4.3.3-2.3.1" class="pilcrow"></a></p>
</li>
<li id="section-4.3.3-2.4">
<p id="section-4.3.3-2.4.1">Temporal Variation: Repeated benchmark
runs <span class="bcp14">MUST</span> vary timing to prevent
memoization attacks.<a href="#section-4.3.3-2.4.1" class="pilcrow"></a></p>
</li>
</ol>
</section>
</div>
</section>
</div>
<div id="performance-claims-in-ect">
<section id="section-4.4">
<h3 id="name-performance-claims-in-ect">
<a href="#section-4.4" class="section-number selfRef">4.4. </a><a href="#name-performance-claims-in-ect" class="section-name selfRef">Performance Claims in ECT</a>
</h3>
<p id="section-4.4-1">Agent ECTs <span class="bcp14">MAY</span> include performance attestation
claims in the <code>ext</code> field:<a href="#section-4.4-1" class="pilcrow"></a></p>
<span class="break"></span><dl class="dlParallel" id="section-4.4-2">
<dt id="section-4.4-2.1">
<code>perf_profile</code>:</dt>
<dd style="margin-left: 1.5em" id="section-4.4-2.2">
<p id="section-4.4-2.2.1">The benchmark profile identifier.<a href="#section-4.4-2.2.1" class="pilcrow"></a></p>
</dd>
<dd class="break"></dd>
<dt id="section-4.4-2.3">
<code>perf_score</code>:</dt>
<dd style="margin-left: 1.5em" id="section-4.4-2.4">
<p id="section-4.4-2.4.1">The aggregate benchmark score.<a href="#section-4.4-2.4.1" class="pilcrow"></a></p>
</dd>
<dd class="break"></dd>
<dt id="section-4.4-2.5">
<code>perf_timestamp</code>:</dt>
<dd style="margin-left: 1.5em" id="section-4.4-2.6">
<p id="section-4.4-2.6.1">The time of the benchmark execution.<a href="#section-4.4-2.6.1" class="pilcrow"></a></p>
</dd>
<dd class="break"></dd>
<dt id="section-4.4-2.7">
<code>perf_harness</code>:</dt>
<dd style="margin-left: 1.5em" id="section-4.4-2.8">
<p id="section-4.4-2.8.1">Identifier of the test harness that produced
the results.<a href="#section-4.4-2.8.1" class="pilcrow"></a></p>
</dd>
<dd class="break"></dd>
</dl>
<p id="section-4.4-3">These claims allow relying parties to evaluate
agent capability before delegation.<a href="#section-4.4-3" class="pilcrow"></a></p>
</section>
</div>
</section>
</div>
<div id="integration-with-ect">
<section id="section-5">
<h2 id="name-integration-with-ect">
<a href="#section-5" class="section-number selfRef">5. </a><a href="#name-integration-with-ect" class="section-name selfRef">Integration with ECT</a>
</h2>
<p id="section-5-1">Behavioral Evidence Tokens integrate into the
ECT DAG defined in
<span>[<a href="#I-D.nennemann-agent-dag-hitl-safety" class="cite xref">I-D.nennemann-agent-dag-hitl-safety</a>]</span> as
follows:<a href="#section-5-1" class="pilcrow"></a></p>
<ol start="1" type="1" class="normal type-1" id="section-5-2">
<li id="section-5-2.1">
<p id="section-5-2.1.1">Each BET references the ECT of the agent
whose behavior was verified via the <code>sub</code>
claim.<a href="#section-5-2.1.1" class="pilcrow"></a></p>
</li>
<li id="section-5-2.2">
<p id="section-5-2.2.1">BETs are attached as child nodes in the
ECT DAG, linked to the agent's execution
node.<a href="#section-5-2.2.1" class="pilcrow"></a></p>
</li>
<li id="section-5-2.3">
<p id="section-5-2.3.1">When an agent delegates to a sub-agent,
the delegating agent's BET chain includes
evidence covering the delegation decision.<a href="#section-5-2.3.1" class="pilcrow"></a></p>
</li>
<li id="section-5-2.4">
<p id="section-5-2.4.1">Verifiers traversing the DAG can inspect
BETs at each node to assess behavioral
compliance across the entire execution
chain.<a href="#section-5-2.4.1" class="pilcrow"></a></p>
</li>
</ol>
<span id="name-bet-integration-in-ect-dag"></span><div id="fig-dag">
<figure id="figure-5">
<div class="alignLeft art-text artwork" id="section-5-3.1">
<pre>
+----------+ +----------+
| ECT |----&gt;| ECT |
| Agent A | | Agent B |
+----+-----+ +----+-----+
| |
+----v-----+ +----v-----+
| BET | | BET |
| Agent A | | Agent B |
+----------+ +----------+
</pre>
</div>
<figcaption><a href="#figure-5" class="selfRef">Figure 5</a>:
<a href="#name-bet-integration-in-ect-dag" class="selfRef">BET Integration in ECT DAG</a>
</figcaption></figure>
</div>
<p id="section-5-4">This structure enables end-to-end behavioral
verification across multi-agent workflows.<a href="#section-5-4" class="pilcrow"></a></p>
</section>
</div>
<div id="security-considerations">
<section id="section-6">
<h2 id="name-security-considerations">
<a href="#section-6" class="section-number selfRef">6. </a><a href="#name-security-considerations" class="section-name selfRef">Security Considerations</a>
</h2>
<div id="adversarial-behavior">
<section id="section-6.1">
<h3 id="name-adversarial-behavior">
<a href="#section-6.1" class="section-number selfRef">6.1. </a><a href="#name-adversarial-behavior" class="section-name selfRef">Adversarial Behavior</a>
</h3>
<p id="section-6.1-1">Agents <span class="bcp14">MAY</span> attempt to behave correctly only when
they detect monitoring. Mitigations include:<a href="#section-6.1-1" class="pilcrow"></a></p>
<ul class="normal">
<li class="normal" id="section-6.1-2.1">
<p id="section-6.1-2.1.1">Unpredictable monitoring intervals<a href="#section-6.1-2.1.1" class="pilcrow"></a></p>
</li>
<li class="normal" id="section-6.1-2.2">
<p id="section-6.1-2.2.1">Covert observation modes where the agent is
not informed of monitor presence<a href="#section-6.1-2.2.1" class="pilcrow"></a></p>
</li>
<li class="normal" id="section-6.1-2.3">
<p id="section-6.1-2.3.1">Cross-referencing BETs with external audit
logs<a href="#section-6.1-2.3.1" class="pilcrow"></a></p>
</li>
</ul>
</section>
</div>
<div id="monitor-compromise">
<section id="section-6.2">
<h3 id="name-monitor-compromise">
<a href="#section-6.2" class="section-number selfRef">6.2. </a><a href="#name-monitor-compromise" class="section-name selfRef">Monitor Compromise</a>
</h3>
<p id="section-6.2-1">A compromised Runtime Monitor could produce
fraudulent BETs. Mitigations include:<a href="#section-6.2-1" class="pilcrow"></a></p>
<ul class="normal">
<li class="normal" id="section-6.2-2.1">
<p id="section-6.2-2.1.1">Monitor attestation using RATS <span>[<a href="#RFC9334" class="cite xref">RFC9334</a>]</span><a href="#section-6.2-2.1.1" class="pilcrow"></a></p>
</li>
<li class="normal" id="section-6.2-2.2">
<p id="section-6.2-2.2.1">Multiple independent monitors with
cross-validation<a href="#section-6.2-2.2.1" class="pilcrow"></a></p>
</li>
<li class="normal" id="section-6.2-2.3">
<p id="section-6.2-2.3.1">Transparency logs for BETs, aligned with
SCITT <span>[<a href="#I-D.ietf-scitt-architecture" class="cite xref">I-D.ietf-scitt-architecture</a>]</span><a href="#section-6.2-2.3.1" class="pilcrow"></a></p>
</li>
</ul>
</section>
</div>
<div id="benchmark-manipulation">
<section id="section-6.3">
<h3 id="name-benchmark-manipulation">
<a href="#section-6.3" class="section-number selfRef">6.3. </a><a href="#name-benchmark-manipulation" class="section-name selfRef">Benchmark Manipulation</a>
</h3>
<p id="section-6.3-1">Agents or their operators <span class="bcp14">MAY</span> attempt to
manipulate benchmark results. The anti-gaming
provisions in Section 4.3.3 address this risk.
Additionally:<a href="#section-6.3-1" class="pilcrow"></a></p>
<ul class="normal">
<li class="normal" id="section-6.3-2.1">
<p id="section-6.3-2.1.1">Benchmark harnesses <span class="bcp14">MUST</span> be operated by
independent parties.<a href="#section-6.3-2.1.1" class="pilcrow"></a></p>
</li>
<li class="normal" id="section-6.3-2.2">
<p id="section-6.3-2.2.1">Results <span class="bcp14">MUST</span> be signed by the harness
operator.<a href="#section-6.3-2.2.1" class="pilcrow"></a></p>
</li>
<li class="normal" id="section-6.3-2.3">
<p id="section-6.3-2.3.1">Benchmark profiles <span class="bcp14">MUST</span> be versioned and
immutable once published.<a href="#section-6.3-2.3.1" class="pilcrow"></a></p>
</li>
</ul>
</section>
</div>
<div id="privacy-of-behavioral-evidence">
<section id="section-6.4">
<h3 id="name-privacy-of-behavioral-evide">
<a href="#section-6.4" class="section-number selfRef">6.4. </a><a href="#name-privacy-of-behavioral-evide" class="section-name selfRef">Privacy of Behavioral Evidence</a>
</h3>
<p id="section-6.4-1">BETs contain information about agent actions
that may be sensitive. Implementations <span class="bcp14">MUST</span>:<a href="#section-6.4-1" class="pilcrow"></a></p>
<ul class="normal">
<li class="normal" id="section-6.4-2.1">
<p id="section-6.4-2.1.1">Minimize the detail in <code>bhv_evidence</code> to
what is necessary for verification.<a href="#section-6.4-2.1.1" class="pilcrow"></a></p>
</li>
<li class="normal" id="section-6.4-2.2">
<p id="section-6.4-2.2.1">Support selective disclosure where possible.<a href="#section-6.4-2.2.1" class="pilcrow"></a></p>
</li>
<li class="normal" id="section-6.4-2.3">
<p id="section-6.4-2.3.1">Protect BETs in transit using TLS
(<span>[<a href="#RFC9110" class="cite xref">RFC9110</a>]</span>).<a href="#section-6.4-2.3.1" class="pilcrow"></a></p>
</li>
<li class="normal" id="section-6.4-2.4">
<p id="section-6.4-2.4.1">Define retention policies for behavioral
evidence.<a href="#section-6.4-2.4.1" class="pilcrow"></a></p>
</li>
</ul>
</section>
</div>
</section>
</div>
<div id="iana-considerations">
<section id="section-7">
<h2 id="name-iana-considerations">
<a href="#section-7" class="section-number selfRef">7. </a><a href="#name-iana-considerations" class="section-name selfRef">IANA Considerations</a>
</h2>
<div id="ect-extension-claim-keys">
<section id="section-7.1">
<h3 id="name-ect-extension-claim-keys">
<a href="#section-7.1" class="section-number selfRef">7.1. </a><a href="#name-ect-extension-claim-keys" class="section-name selfRef">ECT Extension Claim Keys</a>
</h3>
<p id="section-7.1-1">This document requests registration of the
following claim keys in the ECT <code>ext</code> claims
registry:<a href="#section-7.1-1" class="pilcrow"></a></p>
<span id="name-ect-extension-claims-for-be"></span><div id="tbl-claims">
<table class="center" id="table-1">
<caption>
<a href="#table-1" class="selfRef">Table 1</a>:
<a href="#name-ect-extension-claims-for-be" class="selfRef">ECT Extension Claims for Behavioral Verification</a>
</caption>
<thead>
<tr>
<th class="text-left" rowspan="1" colspan="1">Claim Key</th>
<th class="text-left" rowspan="1" colspan="1">Description</th>
</tr>
</thead>
<tbody>
<tr>
<td class="text-left" rowspan="1" colspan="1">bhv_policy</td>
<td class="text-left" rowspan="1" colspan="1">Policy URI reference</td>
</tr>
<tr>
<td class="text-left" rowspan="1" colspan="1">bhv_result</td>
<td class="text-left" rowspan="1" colspan="1">Verification result</td>
</tr>
<tr>
<td class="text-left" rowspan="1" colspan="1">bhv_evidence</td>
<td class="text-left" rowspan="1" colspan="1">Observed actions hash</td>
</tr>
<tr>
<td class="text-left" rowspan="1" colspan="1">bhv_window</td>
<td class="text-left" rowspan="1" colspan="1">Observation period</td>
</tr>
<tr>
<td class="text-left" rowspan="1" colspan="1">bhv_details</td>
<td class="text-left" rowspan="1" colspan="1">Per-behavior results</td>
</tr>
<tr>
<td class="text-left" rowspan="1" colspan="1">perf_profile</td>
<td class="text-left" rowspan="1" colspan="1">Benchmark profile ID</td>
</tr>
<tr>
<td class="text-left" rowspan="1" colspan="1">perf_score</td>
<td class="text-left" rowspan="1" colspan="1">Aggregate benchmark score</td>
</tr>
<tr>
<td class="text-left" rowspan="1" colspan="1">perf_timestamp</td>
<td class="text-left" rowspan="1" colspan="1">Benchmark execution time</td>
</tr>
<tr>
<td class="text-left" rowspan="1" colspan="1">perf_harness</td>
<td class="text-left" rowspan="1" colspan="1">Test harness identifier</td>
</tr>
</tbody>
</table>
</div>
</section>
</div>
<div id="benchmark-profile-media-type">
<section id="section-7.2">
<h3 id="name-benchmark-profile-media-typ">
<a href="#section-7.2" class="section-number selfRef">7.2. </a><a href="#name-benchmark-profile-media-typ" class="section-name selfRef">Benchmark Profile Media Type</a>
</h3>
<p id="section-7.2-1">This document requests registration of the
following media type:<a href="#section-7.2-1" class="pilcrow"></a></p>
<p id="section-7.2-2">Type name: application<a href="#section-7.2-2" class="pilcrow"></a></p>
<p id="section-7.2-3">Subtype name: agent-benchmark-profile+json<a href="#section-7.2-3" class="pilcrow"></a></p>
<p id="section-7.2-4">Required parameters: N/A<a href="#section-7.2-4" class="pilcrow"></a></p>
<p id="section-7.2-5">Optional parameters: N/A<a href="#section-7.2-5" class="pilcrow"></a></p>
<p id="section-7.2-6">Encoding considerations: binary (UTF-8 JSON)<a href="#section-7.2-6" class="pilcrow"></a></p>
<p id="section-7.2-7">Security considerations: See Section 6<a href="#section-7.2-7" class="pilcrow"></a></p>
</section>
</div>
</section>
</div>
<div id="sec-combined-references">
<section id="section-8">
<h2 id="name-references">
<a href="#section-8" class="section-number selfRef">8. </a><a href="#name-references" class="section-name selfRef">References</a>
</h2>
<div id="sec-normative-references">
<section id="section-8.1">
<h3 id="name-normative-references">
<a href="#section-8.1" class="section-number selfRef">8.1. </a><a href="#name-normative-references" class="section-name selfRef">Normative References</a>
</h3>
<dl class="references">
<dt id="I-D.nennemann-agent-dag-hitl-safety">[I-D.nennemann-agent-dag-hitl-safety]</dt>
<dd>
<span class="refTitle">"Agent Context Policy Token: DAG Delegation with Human Override"</span>, <span>n.d.</span>, <span>&lt;<a href="https://datatracker.ietf.org/doc/draft-nennemann-agent-dag-hitl-safety/">https://datatracker.ietf.org/doc/draft-nennemann-agent-dag-hitl-safety/</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="I-D.nennemann-wimse-ect">[I-D.nennemann-wimse-ect]</dt>
<dd>
<span class="refTitle">"Execution Context Tokens for Distributed Agentic Workflows"</span>, <span>n.d.</span>, <span>&lt;<a href="https://datatracker.ietf.org/doc/draft-nennemann-wimse-ect/">https://datatracker.ietf.org/doc/draft-nennemann-wimse-ect/</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC2119">[RFC2119]</dt>
<dd>
<span class="refAuthor">Bradner, S.</span>, <span class="refTitle">"Key words for use in RFCs to Indicate Requirement Levels"</span>, <span class="seriesInfo">BCP 14</span>, <span class="seriesInfo">RFC 2119</span>, <span class="seriesInfo">DOI 10.17487/RFC2119</span>, <time datetime="1997-03" class="refDate">March 1997</time>, <span>&lt;<a href="https://www.rfc-editor.org/rfc/rfc2119">https://www.rfc-editor.org/rfc/rfc2119</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC7515">[RFC7515]</dt>
<dd>
<span class="refAuthor">Jones, M.</span>, <span class="refAuthor">Bradley, J.</span>, and <span class="refAuthor">N. Sakimura</span>, <span class="refTitle">"JSON Web Signature (JWS)"</span>, <span class="seriesInfo">RFC 7515</span>, <span class="seriesInfo">DOI 10.17487/RFC7515</span>, <time datetime="2015-05" class="refDate">May 2015</time>, <span>&lt;<a href="https://www.rfc-editor.org/rfc/rfc7515">https://www.rfc-editor.org/rfc/rfc7515</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC7519">[RFC7519]</dt>
<dd>
<span class="refAuthor">Jones, M.</span>, <span class="refAuthor">Bradley, J.</span>, and <span class="refAuthor">N. Sakimura</span>, <span class="refTitle">"JSON Web Token (JWT)"</span>, <span class="seriesInfo">RFC 7519</span>, <span class="seriesInfo">DOI 10.17487/RFC7519</span>, <time datetime="2015-05" class="refDate">May 2015</time>, <span>&lt;<a href="https://www.rfc-editor.org/rfc/rfc7519">https://www.rfc-editor.org/rfc/rfc7519</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC8174">[RFC8174]</dt>
<dd>
<span class="refAuthor">Leiba, B.</span>, <span class="refTitle">"Ambiguity of Uppercase vs Lowercase in RFC 2119 Key Words"</span>, <span class="seriesInfo">BCP 14</span>, <span class="seriesInfo">RFC 8174</span>, <span class="seriesInfo">DOI 10.17487/RFC8174</span>, <time datetime="2017-05" class="refDate">May 2017</time>, <span>&lt;<a href="https://www.rfc-editor.org/rfc/rfc8174">https://www.rfc-editor.org/rfc/rfc8174</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC9334">[RFC9334]</dt>
<dd>
<span class="refAuthor">Birkholz, H.</span>, <span class="refAuthor">Thaler, D.</span>, <span class="refAuthor">Richardson, M.</span>, <span class="refAuthor">Smith, N.</span>, and <span class="refAuthor">W. Pan</span>, <span class="refTitle">"Remote ATtestation procedureS (RATS) Architecture"</span>, <span class="seriesInfo">RFC 9334</span>, <span class="seriesInfo">DOI 10.17487/RFC9334</span>, <time datetime="2023-01" class="refDate">January 2023</time>, <span>&lt;<a href="https://www.rfc-editor.org/rfc/rfc9334">https://www.rfc-editor.org/rfc/rfc9334</a>&gt;</span>. </dd>
<dd class="break"></dd>
</dl>
</section>
</div>
<div id="sec-informative-references">
<section id="section-8.2">
<h3 id="name-informative-references">
<a href="#section-8.2" class="section-number selfRef">8.2. </a><a href="#name-informative-references" class="section-name selfRef">Informative References</a>
</h3>
<dl class="references">
<dt id="I-D.ietf-scitt-architecture">[I-D.ietf-scitt-architecture]</dt>
<dd>
<span class="refAuthor">Birkholz, H.</span>, <span class="refAuthor">Delignat-Lavaud, A.</span>, <span class="refAuthor">Fournet, C.</span>, <span class="refAuthor">Deshpande, Y.</span>, and <span class="refAuthor">S. Lasker</span>, <span class="refTitle">"An Architecture for Trustworthy and Transparent Digital Supply Chains"</span>, <span class="refContent">Work in Progress</span>, <span class="seriesInfo">Internet-Draft, draft-ietf-scitt-architecture-22</span>, <time datetime="2025-10-10" class="refDate">10 October 2025</time>, <span>&lt;<a href="https://datatracker.ietf.org/doc/html/draft-ietf-scitt-architecture-22">https://datatracker.ietf.org/doc/html/draft-ietf-scitt-architecture-22</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="I-D.nennemann-agent-gap-analysis">[I-D.nennemann-agent-gap-analysis]</dt>
<dd>
<span class="refTitle">"Gap Analysis for Autonomous Agent Protocols"</span>, <span>n.d.</span>, <span>&lt;<a href="https://datatracker.ietf.org/doc/draft-nennemann-agent-gap-analysis/">https://datatracker.ietf.org/doc/draft-nennemann-agent-gap-analysis/</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC9110">[RFC9110]</dt>
<dd>
<span class="refAuthor">Fielding, R., Ed.</span>, <span class="refAuthor">Nottingham, M., Ed.</span>, and <span class="refAuthor">J. Reschke, Ed.</span>, <span class="refTitle">"HTTP Semantics"</span>, <span class="seriesInfo">STD 97</span>, <span class="seriesInfo">RFC 9110</span>, <span class="seriesInfo">DOI 10.17487/RFC9110</span>, <time datetime="2022-06" class="refDate">June 2022</time>, <span>&lt;<a href="https://www.rfc-editor.org/rfc/rfc9110">https://www.rfc-editor.org/rfc/rfc9110</a>&gt;</span>. </dd>
<dd class="break"></dd>
</dl>
</section>
</div>
</section>
</div>
<div id="acknowledgments">
<section id="appendix-A">
<h2 id="name-acknowledgments">
<a href="#name-acknowledgments" class="section-name selfRef">Acknowledgments</a>
</h2>
<p id="appendix-A-1">The author thanks the contributors to the NMOP
working group for discussions on agent
operational requirements.<a href="#appendix-A-1" class="pilcrow"></a></p>
</section>
</div>
<div id="authors-addresses">
<section id="appendix-B">
<h2 id="name-authors-address">
<a href="#name-authors-address" class="section-name selfRef">Author's Address</a>
</h2>
<address class="vcard">
<div dir="auto" class="left"><span class="fn nameRole">Christian Nennemann</span></div>
<div dir="auto" class="left"><span class="org">Independent Researcher</span></div>
<div class="email">
<span>Email:</span>
<a href="mailto:ietf@nennemann.de" class="email">ietf@nennemann.de</a>
</div>
</address>
</section>
</div>
<script>const toc = document.getElementById("toc");
toc.querySelector("h2").addEventListener("click", e => {
toc.classList.toggle("active");
});
toc.querySelector("nav").addEventListener("click", e => {
toc.classList.remove("active");
});
</script>
</body>
</html>