{"id":995,"date":"2026-11-20T06:51:03","date_gmt":"2026-11-20T06:51:03","guid":{"rendered":"https:\/\/ont.io\/news\/?p=995"},"modified":"2026-11-20T06:51:30","modified_gmt":"2026-11-20T06:51:30","slug":"ai-agent-identity-verified-consent","status":"publish","type":"post","link":"https:\/\/ont.io\/news\/ai-agent-identity-verified-consent\/","title":{"rendered":"AI agent identity: what a service must check before an agent acts"},"content":{"rendered":"<p>On 16 November, Federal Reserve Governor Michael Barr told a conference in Milan that major banks now report finding AI agents attempting to take actions in retail customer accounts, &#8220;without the bank being able to verify authenticity, consent, and alignment with customer instructions&#8221; (<a href=\"https:\/\/www.federalreserve.gov\/newsevents\/speech\/barr20261116a.htm\" target=\"_blank\" rel=\"noopener\">Federal Reserve, 16 November 2026<\/a>). Each of those is a question of AI agent identity: who the agent acts for, what it has been allowed to do, and whether that permission still stands.<\/p>\n<p>A day later, Poland&#8217;s Council of Ministers adopted the draft law that will let it launch the European Digital Identity Wallet, and the government says users will decide for themselves which data each service receives (<a href=\"https:\/\/www.gov.pl\/web\/premier\/projekt-ustawy-o-zmianie-ustawy-o-uslugach-zaufania-oraz-identyfikacji-elektronicznej-oraz-niektorych-innych-ustaw3\" target=\"_blank\" rel=\"noopener\">Chancellery of the Prime Minister, 17 November 2026<\/a>). In both cases a service has to check what a person authorised before it acts. AI agent identity adds one more question: whether the software presenting that permission is the software it was given to.<\/p>\n<h2>Banks report AI agents acting without verified consent<\/h2>\n<p>His list of guardrails was &#8220;strict authentication, explicit consent verification for specific actions, technical controls limiting agent capabilities, and both rules- and values-based constraints&#8221; (<a href=\"https:\/\/www.federalreserve.gov\/newsevents\/speech\/barr20261116a.htm\" target=\"_blank\" rel=\"noopener\">Federal Reserve, 16 November 2026<\/a>).<\/p>\n<p>A bank can authenticate its own customer at login. An agent calling the account on that customer&#8217;s behalf arrives with none of that history, so the bank has to decide in the moment whether the customer granted this agent this permission, for this action, and whether it has since been withdrawn. A credential can carry each of those claims.<\/p>\n<h2>Four checks AI agent identity has to pass<\/h2>\n<p><strong>Who granted the permission.<\/strong> The consent has to be signed by an identifier the customer controls, with public keys the bank can look up.<\/p>\n<p><strong>What was granted.<\/strong> A specific action, a limit and a time window. The W3C Verifiable Credentials Data Model 2.0 defines validity periods, so a permission can carry its own start and end (<a href=\"https:\/\/www.w3.org\/TR\/vc-data-model-2.0\/\" target=\"_blank\" rel=\"noopener\">W3C Recommendation, 15 May 2025<\/a>).<\/p>\n<p><strong>That the permission still stands.<\/strong> If the customer withdraws consent, every later check has to fail. The W3C Bitstring Status List standard gives verifiers a way to read revocation status (<a href=\"https:\/\/www.w3.org\/TR\/vc-bitstring-status-list\/\" target=\"_blank\" rel=\"noopener\">W3C Recommendation, 15 May 2025<\/a>).<\/p>\n<p><strong>That the agent presenting it is the one it was given to.<\/strong> A copied permission should be worthless. Under the W3C DID standard, the controller of an identifier can prove control over it &#8220;without requiring permission from any other party&#8221; (<a href=\"https:\/\/www.w3.org\/TR\/did-core\/\" target=\"_blank\" rel=\"noopener\">W3C Recommendation, 19 July 2022<\/a>).<\/p>\n<h2>Ontology&#8217;s chain answers who issued it, whether it stands and who holds it<\/h2>\n<p>ONT ID is Ontology&#8217;s decentralized identity framework, built on the W3C Decentralized Identifiers model. The holder controls the keys, keys can be rotated without changing the identifier, and one holder can keep several identifiers for different purposes (<a href=\"https:\/\/ont.io\/news\/decentralized-identity-how-it-works\/\">Ontology, 27 October 2026<\/a>). In the W3C model a DID can refer to any subject, &#8220;a person, organization, thing, data model, abstract entity&#8221; (<a href=\"https:\/\/www.w3.org\/TR\/did-core\/\" target=\"_blank\" rel=\"noopener\">W3C, 19 July 2022<\/a>), so the same identifier model can serve AI agent identity.<\/p>\n<p>What Ontology writes to its public chain is deliberately small: the issuer&#8217;s identifier and keys so a verifier can check the signature, the holder&#8217;s identifier so the credential can be tied to whoever presents it, and a status list entry so the issuer can revoke it. The credential&#8217;s content stays off chain with the holder (<a href=\"https:\/\/ont.io\/news\/decentralized-identity-how-it-works\/\">Ontology, 27 October 2026<\/a>).<\/p>\n<p>From those records a verifier can confirm who issued a credential, that it has not been revoked, and that the holder presenting it controls the identifier it was issued to, all by reading Ontology&#8217;s public chain and without contacting the issuer (<a href=\"https:\/\/ont.io\/news\/portable-reputation-belongs-to-the-person\/\">Ontology, 4 November 2026<\/a>). The status list lets every later verifier see a withdrawal without the issuer learning who checked. Because the chain is a shared, append-only record, an issuer cannot quietly backdate or reorder events (<a href=\"https:\/\/ont.io\/news\/portable-reputation-belongs-to-the-person\/\">Ontology, 4 November 2026<\/a>). That property matters when a dispute turns on whether a permission was withdrawn before or after an action.<\/p>\n<h2>A consent credential for an AI agent, step by step<\/h2>\n<p>This pattern uses only parts of the model that exist today: identifiers, signed credentials, validity periods and status lists.<\/p>\n<ol>\n<li>The customer holds an identifier and its keys.<\/li>\n<li>The agent is given its own identifier, controlled by keys that its operator holds.<\/li>\n<li>The customer signs a credential that names the agent&#8217;s identifier and states the permitted action, the limit and the expiry. An issuer is anyone whose word a verifier will accept (<a href=\"https:\/\/ont.io\/news\/decentralized-identity-how-it-works\/\">Ontology, 27 October 2026<\/a>).<\/li>\n<li>The agent presents the credential and signs the request, proving it controls the identifier the credential names.<\/li>\n<li>The bank checks the customer&#8217;s signature against keys on the chain, reads the status list, confirms the validity window and confirms the agent&#8217;s control. Then it applies its own limits and rules.<\/li>\n<li>To withdraw consent, the customer revokes the credential, and the next check fails.<\/li>\n<\/ol>\n<p>This gives the bank checkable evidence of who authorised what. Barr&#8217;s other guardrails, technical limits on what an agent can do and rules for how it behaves, stay with the bank and with whoever builds the agent.<\/p>\n<h2>Europe&#8217;s identity wallets use the same model<\/h2>\n<p>Under the amended eIDAS regulation, every EU member state has to make at least one European Digital Identity Wallet available by the end of 2026 (<a href=\"https:\/\/ont.io\/news\/decentralized-identity-how-it-works\/\">Ontology, 27 October 2026<\/a>). Poland&#8217;s wallet will first run as a separate app reachable from mObywatel, and the two will later be combined (<a href=\"https:\/\/www.gov.pl\/web\/premier\/projekt-ustawy-o-zmianie-ustawy-o-uslugach-zaufania-oraz-identyfikacji-elektronicznej-oraz-niektorych-innych-ustaw3\" target=\"_blank\" rel=\"noopener\">Chancellery of the Prime Minister, 17 November 2026<\/a>). The pilot version will require a one-time strong authentication with an electronic identity card before the wallet can be used (<a href=\"https:\/\/www.intellinews.com\/poland-to-link-eu-digital-identity-wallet-with-mobywatel-app-486248\/\" target=\"_blank\" rel=\"noopener\">bne IntelliNews, 18 November 2026<\/a>).<\/p>\n<p>That is the holder, issuer and verifier model inside national trust frameworks. Ontology runs the same model on a public chain that works across borders and industries, and has done since 2018 (<a href=\"https:\/\/ont.io\/news\/decentralized-identity-how-it-works\/\">Ontology, 27 October 2026<\/a>).<\/p>\n<h2>ONTO Wallet&#8217;s gaming reputation uses the same checks<\/h2>\n<p>On 19 November ONTO Wallet set out why it is building a gaming reputation that players hold in their wallets, tied to their ONT ID. Players share it item by item, for a limited time, and can withdraw it. A tournament or guild can check who issued it, that it has not been withdrawn and that it belongs to the player by reading Ontology&#8217;s public chain. The gaming reputation is in development (<a href=\"https:\/\/onto.app\/blog\/gaming-reputation-you-hold\/\" target=\"_blank\" rel=\"noopener\">ONTO Wallet, 19 November 2026<\/a>).<\/p>\n<p>A tournament checking a player and a bank checking an agent ask the same three questions: who issued this, is it still valid, and does it belong to whoever is presenting it.<\/p>\n<p>If you are building a service that agents will call and need AI agent identity you can verify, Ontology&#8217;s identity layer is ready to build on.<\/p>\n<h2>Common questions about AI agent identity<\/h2>\n<h3>Can an AI agent have an ONT ID?<\/h3>\n<p>ONT ID follows the W3C DID model, in which an identifier can refer to any subject, including a thing (<a href=\"https:\/\/www.w3.org\/TR\/did-core\/\" target=\"_blank\" rel=\"noopener\">W3C, 19 July 2022<\/a>). Ontology&#8217;s developer documentation describes ONT ID identities for individuals, institutions, objects and content (<a href=\"https:\/\/docs.ont.io\/decentralized-identity-and-data\/ontid\">Ontology docs<\/a>). Whoever holds the keys controls the identifier.<\/p>\n<h3>Does this replace a bank&#8217;s own authentication?<\/h3>\n<p>No. It gives the bank evidence it can check about who granted a permission and whether it still stands. The bank&#8217;s own authentication, limits and controls still apply.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Banks report AI agents trying to act in customer accounts without verified consent. Four checks a service needs, and how Ontology&#8217;s chain answers them.<\/p>\n","protected":false},"author":5,"featured_media":997,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[170,13],"tags":[36],"class_list":["post-995","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai","category-did-and-privacy","tag-ontology-network"],"_links":{"self":[{"href":"https:\/\/ont.io\/news\/wp-json\/wp\/v2\/posts\/995","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/ont.io\/news\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/ont.io\/news\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/ont.io\/news\/wp-json\/wp\/v2\/users\/5"}],"replies":[{"embeddable":true,"href":"https:\/\/ont.io\/news\/wp-json\/wp\/v2\/comments?post=995"}],"version-history":[{"count":2,"href":"https:\/\/ont.io\/news\/wp-json\/wp\/v2\/posts\/995\/revisions"}],"predecessor-version":[{"id":999,"href":"https:\/\/ont.io\/news\/wp-json\/wp\/v2\/posts\/995\/revisions\/999"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/ont.io\/news\/wp-json\/wp\/v2\/media\/997"}],"wp:attachment":[{"href":"https:\/\/ont.io\/news\/wp-json\/wp\/v2\/media?parent=995"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/ont.io\/news\/wp-json\/wp\/v2\/categories?post=995"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/ont.io\/news\/wp-json\/wp\/v2\/tags?post=995"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}