Skip to content
IAM Insights

Explainers · Glossary

CAEP

Continuous Access Evaluation Profile

By Boston Identity · Explained · 6 min read

The short answer

CAEP is an OpenID standard for the events one system sends another after login, when a session is revoked, a credential changes or a device falls out of compliance, so the receiver can cut back access.

What it is
A standard
Defined by
OpenID Foundation, CAEP 1.0 Final
Who it's for
Teams that must end access fast across apps
In our briefings
29By month of the source: 2025-12: 0, 2026-01: 1, 2026-02: 1, 2026-03: 1, 2026-04: 2, 2026-05: 1, 2026-06: 1, 2026-07: 4, 2026-08: 3, 2026-09: 7, 2026-10: 3, 2026-11: 00by month, last 12
01

What is CAEP?

The Continuous Access Evaluation Profile (CAEP) is a set of event types that one system sends to another.§1 It is a profile of the OpenID Shared Signals Framework (SSF): the events it defines conform to SSF.§1 The sender is the transmitter and the system that listens is the receiver. The spec's aim is that a receiver can cut back access to the users, devices, sessions and applications the two of them share.S1

CAEP deals with the time after login. The OpenID Foundation's release says sessions often last days or weeks while a user's device compliance or access can change, and that systems relying on federated identity had no way to hear about it.1 CAEP 1.0 defines eight events: Session Revoked, Token Claims Change, Credential Change, Assurance Level Change, Device Compliance Change, Risk Level Change, Session Established and Session Presented.§3 The last two report where a user's sessions are open. An identity provider can use Session Established to detect unintended logins, and a receiver can use either event to build an inventory of a user's sessions.§3.6,§3.7

CAEP is not a policy engine. It tells the receiver what changed, but for most events it says almost nothing about what the receiver must do next.2

02

How does CAEP work?

Two parties take part. The transmitter notices a change, and the receiver hears about it and decides what to do. In the spec's examples the transmitter is an identity provider and the receiver is an application, though a service can act as a transmitter too.§3.6

Figure steps
  1. Two systems agree to share signals
  2. Something changes at the transmitter
  3. The transmitter names who it is about
  4. The transmitter sends the event with its details
  5. The receiver decides how to respond
Fig. 1

How a CAEP event moves from the transmitter that sees a change to the receiver that acts on it.

Drawn from OpenID CAEP 1.0, sections 1–4

Swipe for the full figure →

  1. Two systems agree to share signals

    A transmitter and a receiver agree to cooperate. CAEP rides on SSF, and the spec says implementations should comply with SSF.§4 CAEP's own job is to define the event types that travel between them.§1

  2. Something changes at the transmitter

    A session is revoked, a user resets a password or enrolls a new second factor, a device drops out of compliance, or a risk level changes.§3.1,§3.3,§3.5,§3.8 Each kind of change has its own event type and a fixed name under the CAEP base address.§3

  3. The transmitter names who it is about

    The event's subject can point straight at one session, or it can combine a user, a device and a tenant into a complex subject.§3.1 When a complex subject is used, a Session Revoked event applies to any session that matches those combined claims.§3.1

  4. The transmitter sends the event with its details

    The event travels as a token addressed to the receiver, with the subject and the event inside.§3.1.2 Some types carry their own required fields. A Credential Change, for example, must say the credential type and whether it was created, revoked, updated or deleted.§3.3.1 Any event may add when it happened, whether an admin, user, policy or system triggered it, and a reason for the log or for the user.§2

  5. The receiver decides how to respond

    The spec's purpose is that receivers can cut back access, but it does not say how.S1 Session Revoked carries no event-specific fields and states no required response or deadline.§3.1.1,2 One receiver rule is spelled out: if an assurance or risk event leaves out the previous level, the receiver must assume the transmitter does not know it.§3.4.1,§3.8.1

03

Why does CAEP matter now?

On 16 September 2025 the OpenID Foundation announced that it had approved CAEP 1.0 as a Final Specification, together with SSF 1.0 and RISC 1.0.1 The release reports no deployment or test result, and its list of adopters names no product.1 Sean O'Dell, who works on the Shared Signals standards and on identity at Disney, recalls more than 30 people testing at one interoperability event, and that is his recollection.3 He also points to Okta as already sending and receiving CAEP events, which he calls public knowledge.4

Buyers now ask vendors about Shared Signals and CAEP by name.5 Two buyers on one panel expect a vendor of tools for AI agents and other non-human identities to speak to CAEP, Shared Signals and transaction tokens.6

Agnihotri argues that a normal session token life of fifteen minutes to an hour is long enough for an autonomous agent to finish its damage.7 He says shared signals and CAEP are the plumbing that could change the habit of answering an incident with a mass credential reset.7 An Internet Engineering Task Force (IETF) draft on AI agents says any participant may subscribe to Shared Signals events under the CAEP or RISC profiles.8,9 The same draft says a receiver must make sure a revocation is enforced without undue delay.8 SailPoint's chief product officer describes a design that kills the session when a shared-signals feed reports the human's account compromised; this is the design as described, not a shipped feature list.10 We forecast that by March 2028 the Shared Signals working group publishes events made for agents, such as an agent's delegation being withdrawn.11

04

Where does it fit?

Compared withHow CAEP differs
SSFSSF is the framework that streams events between a transmitter and a receiver, by push or by poll.12 The only event types SSF defines itself are stream verification and stream updates; session revocation comes from CAEP.13
RISCRISC is the sibling profile, announced as Final alongside CAEP.1 Account-disabled is a RISC event, while session-revoked is a CAEP one.13
SAMLCAEP works beside SAML (Security Assertion Markup Language) logins. A Token Claims Change event can name a SAML assertion as its subject, and Session Established can carry an external id that ties a session to a federated session set up with SAML.§3.2.2,§3.6.1
05

What goes wrong?

  • A receiver can accept an event that its gateways never act on. In a 16-host lab test by Liu, Yu and Zhang, a network rule cut one receiver's handoff to its gateways. The receiver still acknowledged and stored the revocation, and those gateways kept letting calls through.14 The authors say a poll cannot be assumed to repair a pushed event the transmitter already counts as delivered.12
  • Each receiver writes its own rule. CAEP names the change but sets no required response, so what happens next depends on each receiver's policy.2,4
  • A receiver that acts on risk events trusts the transmitter's scoring. A Risk Level Change carries LOW, MEDIUM or HIGH, the signals behind it are outside the event's scope, and a reason is only recommended.§3.8,15
  • Skipping SSF may cause security issues. The spec's only security note says implementations should comply with SSF, and that exchanging the events without it may result in security issues.§4
06

What should a security leader ask?

  1. 01

    When our identity provider revokes a session, which of our applications receive a CAEP Session Revoked event, and which keep working until their tokens expire?

  2. 02

    For each event type we receive, what does each application do, and who wrote that rule down?

  3. 03

    How would we know that a revocation a receiver acknowledged actually reached every gateway behind it?

  4. 04

    Which of our vendors' products send CAEP events and which receive them today, by product and version?

  5. 05

    When a Risk Level Change arrives, whose scoring are we acting on, and can we see the reason behind it?

07

Sources

›Evidence: 15 claims from 12 briefings
  1. 1BI-FIND-2025-0282 · OpenID Foundation finalized Shared Signals, CAEP and RISC in September 2025, naming adopters without evidenceOn 16 September 2025 the OpenID Foundation announced that it had approved three Final Specifications: the Shared Signals Framework 1.0, the Continuous Access Evaluation Profile (CAEP) 1.0 and Risk Information Sharing and Coordination (RISC) 1.0. They are aimed at the time after login. The release says sessions "often last days or weeks" while a user's device compliance or access can change, and that systems relying on federated identity had no way to hear about it. The case it makes for adopting them now rests on its own word: it reports no deployment, interoperability result or implementer count. Its line that "Apple, IBM, Okta, and others, have already adopted these protocols" names no product, version or link, and Executive Director Gail Hodges says Final status "unlocks the ability of many governments to adopt the specifications" without naming a government.
  2. 2BI-FIND-2025-0422 · CAEP 1.0 tells apps what changed after login and leaves almost every response to the receiverCAEP 1.0 defines eight event types and spells out what each carries, but says almost nothing about what a receiver must do with them. Credential Change, for instance, requires a credential type from a list that includes fido2-roaming and verifiable-credential, or one both sides support, and Token Claims Change requires the changed claims with their new values; neither says what the receiver should do next. The only MUST that names the receiver covers a missing previous level in the Assurance Level Change and Risk Level Change events: the receiver "MUST assume that the previous assurance level is unknown to the Transmitter". Session Revoked has no event-specific claims and states no required response or deadline. With a complex subject, the spec says it "applies to any session derived from matching those combined claims". The security considerations say only that implementations SHOULD comply with the Shared Signals Framework (SSF), and that exchanging the events without it may result in security issues.
  3. 3BI-FIND-2025-0470 · Sean O'Dell says Shared Signals audiences have stopped asking what CAEP is and started asking how to deploy itSean O'Dell, who works on the Shared Signals standards as well as on identity at Disney, says the questions changed in two years. In 2023 he had to show audiences what a JWT looks like; at Identiverse 2025 the panel's audience "were looking for prescriptive guidance on it." He recalls more than 30 people testing at the last interoperability event, and Mike Kaiser saying from the stage that he had never seen adoption that fast. Both are his recollections, and he names no tester and no result. He says he asks vendors what their plan to support SSF is, and a host urges customers to ask the same, as a requirement, because "the more customers who do that, the faster it gets out to a roadmap".
  4. 4BI-FIND-2025-0472 · Sean O'Dell says Shared Signals audiences have stopped asking what CAEP is and started asking how to deploy itO'Dell splits the policy between the two ends of a signal. The transmitter's rule decides what goes out: in his example, a malicious session-revoke signal with a risk score over 90 is sent on to the systems it serves. The receiver's rule decides what happens next, such as ending the session whenever its identity provider says to. Where an event goes depends on its domain: provisioning, lifecycle and attribute changes to the governance (IGA) platform, CAEP session events to applications. He points to Okta as already sending and receiving CAEP events, which he calls public knowledge, and says every platform should aspire to be both a transmitter and a receiver. He wants outside sources such as EDR in the mix. Asked whether outside signals can be trusted, his short answer is no: their worth depends on how many other sources sit beside them, and the weight each gets is, in his word, "subjective".
  5. 5BI-FIND-2026-0474 · Eight panels agree an agent acts for a person, never as that person. When agents hand work to other agents, nobody had an answer.The summit wants only a few things from standards, and no existing standard covers any of them. Step-up authorization, where the agent asks for more authority in the middle of a task, has no standard. Nobody is writing assurance levels for agents, the graded levels of trust in an identity. No open standard says how an agent's permissions are set when it is registered. The Model Context Protocol (MCP), with its OAuth layer, is the protocol agents use to call tools, and the Governing AI panel agrees it is not a policy system. The pieces that exist are being stretched to fit. The transaction-token draft is pitched as a way to detect agents, though it never mentions an agent. Intent-based access control, which grants access based on what the agent is trying to do, keeps turning back into ordinary authorization carried in a token. Buyers now ask vendors by name about transaction tokens and about Shared Signals and its Continuous Access Evaluation Profile (CAEP), standards for passing security events such as a revoked session between systems. The vocabulary has reached buyers ahead of the specifications.
  6. 6BI-FIND-2026-0468 · Vendors all claim to cover AI agents and non-human identities, so buyers test them in their own environmentWhat changes when the purchase is an agentic or NHI tool rather than classic IAM is the vetting list, on the account of one panel's two buyers. The incumbent platforms get checked first. A vendor is expected to speak to transaction tokens, CAEP and Shared Signals, identity fabric and intent, and to answer what has to be in place before the tool works, because the buyers' own experience is that they buy the Cadillac and implement the Chevy. The standards that two days earlier the draft authors pitched from the same stage are now a buyer's checklist.
  7. 7BI-FIND-2026-0134 · An AI agent can do its damage within one session, so a practitioner wants a kill switch on its whole identityA session token's normal lifetime, fifteen minutes to an hour, is long enough for an autonomous agent to finish its damage, which is why Agnihotri wants a kill switch on the whole identity rather than a shorter session. Roy Akerman sets up the attacker's version, in and out before behavioural monitoring joins the dots, and Agnihotri adds that most SOC teams still answer an incident with a mass credential reset because they cannot tell what happened; the shared-signals work and CAEP are the plumbing that could change that. What he does not say is which tokens, sessions and grants the switch would have to reach, the gap the corpus's offboarding and block-the-identity threads already hold open.
  8. 8BI-FIND-2026-0182 · Six companies' agent-auth draft, since adopted by the WIMSE group, keeps credentials out of the model's reachThe draft calls observability "a security control, not solely an operational feature". Any participant MAY subscribe to Shared Signals Framework events under the CAEP or RISC profiles. A recipient of a revocation or risk signal SHOULD remediate and MUST ensure revoked or downgraded authorization is enforced without undue delay, and "Cached authorization decisions and tokens that are no longer valid MUST NOT continue to be used after a revocation or risk notification is received". Audit records MUST be tamper-evident and MUST record at least the agent identifier, the delegated subject when there is one, the resource, the action and decision, a timestamp with a correlation identifier, the posture or risk state, and any remediation. The draft does not say how the logs are made tamper-evident, and it leaves the policy model, its format and compliance criteria out of scope as specific to each deployment.
  9. 9BI-FIND-2026-0231 · Containing an agent is not a backup to defining its task; where the agent cannot see everything, it comes firstRuntime alignment, in McGuinness's design, never widens what an agent may do; it decides whether the system still trusts the agent's judgment inside bounds already set. The instrument is a trust budget, which he is careful to call a continued-reliance threshold rather than additional privilege, aggregating drift across a run while never averaging away a single failed check. What makes it work is where the evidence comes from: telemetry attested through workload identity such as SPIFFE, and a Continuous Access Evaluation or Shared Signals path that carries the decision, so that the agent cannot self-report its own good behaviour. The IETF agent draft reaches the same place from the other direction, with every participant subscribing to shared signals and treating observability as a security control.
  10. 10BI-FIND-2026-0077 · SailPoint's product chief argues an agent's access should be re-checked at every action, not once at setupSailPoint's stated enforcement design has three layers: a prompt layer that screens the first user-to-agent request for injection, a pre-action layer that checks the agent's planned tool and MCP calls against the delegating human's entitlements before anything runs, and a runtime layer that authorizes each resource access, narrows scope with OAuth token exchange, and kills the session when a shared-signals feed reports the human's account compromised. This is the design as the chief product officer describes it, not a shipped feature list; the pre-action check is the part no competitor in this corpus has described.
  11. 11BI-CALL-2026-0044 · The panel expects agent lifecycle management to fail first; Eve Maler wants assurance levels for agents in 2027By March 2028 the OpenID Foundation's Shared Signals working group publishes a draft or profile that carries agent-specific events, such as an agent's credential being revoked, its delegation withdrawn, or its session terminated, so an identity provider can push an agent's change of state to the systems it touches. Settles wrong if CAEP and RISC still describe only human sessions and devices by then.
  12. 12BI-FIND-2026-1795 · Receiver acknowledged, gateways cut off: in a fixed 16-host replay, 112 of 225 revoked tool calls still committedThe authors want any API that reports a revocation as complete to say what it checked: the intended gateway set, receipts bound to the revocation and to each gateway, the call population, the observation window and the source of truth for effects. Their monitor answers with one of four states (complete-for-bound, partial-enforcement, indeterminate-evidence or violated), and a recorded violation stands even if a receipt arrives late. In their reading, today's standards stop earlier. Global Token Revocation's completion condition concerns issuer-side token revocation and reauthentication and "does not require receipts from a distributed gateway population", while CAEP and the Shared Signals Framework define events and push or poll delivery between transmitters and receivers. Their Keycloak runs show why a second delivery path is not enough by itself: push and poll each reached gateway enforcement when the handoff completed, but a poll cannot be assumed to repair a push event the transmitter already counts as delivered. So they want repair triggered by missing enforcement receipts, an independently readable event source, or pending state that survives until the gateways enforce.
  13. 13BI-FIND-2025-0151 · Under Shared Signals 1.0 a sender may accept a subscription and send nothing, so agent revocation needs a backupThe framework never mentions agents, workloads or AI. Its subjects are "human or robotic principals, devices, customer tenants in a multi-tenanted service, organizational units within a tenant, groups of subject principals, or other entities", and its complex subject has members for user, device, session, application, tenant, org_unit and group. A complex subject must refer to exactly one principal, which we read as ruling out a single subject that names both an agent and the person it acts for. That pairing could travel as an extra subject member defined by an event type, which the specification allows, or in a proprietary identifier format the parties agree between them. The revocation events themselves come from elsewhere: session-revoked is a CAEP event and account-disabled a RISC one, and the only event types SSF defines itself are stream verification and stream-updated.
  14. 14BI-FIND-2026-1792 · Receiver acknowledged, gateways cut off: in a fixed 16-host replay, 112 of 225 revoked tool calls still committedIn Liu, Yu and Zhang's 16-host test, a revocation that a receiver had acknowledged and stored never reached the gateways behind it, and those gateways kept letting calls through. Keycloak sent the revocation as a Shared Signals event to two receivers, each serving four of eight tool gateways, and a network rule cut only receiver B's handoff to its gateways. Receiver B still answered HTTP 202 and persisted the event. Its gateways admitted every call assigned to them and every one committed: 3,360 calls in each of two separately built executions, 112 of 225 (49.78%) in every block. Normal delivery and recovery committed none. The authors say the 49.78% "should not be read as an incident rate", since it follows from receiver B owning half the gateways. The test ran Keycloak's own delivery code with receivers and gateways the authors wrote, and the paper attributes the gap to no Keycloak defect.
  15. 15BI-FIND-2025-0423 · CAEP 1.0 tells apps what changed after login and leaves almost every response to the receiverSeveral CAEP events give the receiver the transmitter's grade without its basis. Risk Level Change carries a level that "MUST be one of LOW, MEDIUM, HIGH" for a user, device, session, tenant, org unit, group or other principal. The signals behind it come from channels "beyond the scope of this event description", and a risk_reason is only RECOMMENDED. Device Compliance Change reports a device as compliant or not-compliant. Assurance Level Change takes its levels from a namespace that MAY be RFC 8176, RFC 6711, ISO/IEC 29115, NIST's IAL, AAL or FAL, or a custom one the two parties agree. We read this as: a receiver that acts on these events is trusting the transmitter's scoring.

›In our briefings

The claim each of the newest briefings on CAEP would have you leave with.

  1. Asked which identity and security controls are most likely to break first as agentic systems scale, John Yeoh, CSA's chief scientific officer, names lifecycle management. He cites CircleCI and Microsoft's Midnight Blizzard breach as poor lifecycle management of non-human identities, where keys for secrets were not rotated fast enough, and expects agents to fail faster because people now give them specs that spin up sub-agents. Access reviews that ignore intent and behaviour come next, then logging, in what he calls "a good domino of control collapse". Aizome's Amir Ofek answers the same question differently: applying controls built for non-human identities, which are deterministic rather than dynamic, is what will break first.

    CSA's John Yeoh expects lifecycle management to be the first agent control to breakBI-FIND-2026-0113

  2. Agnihotri argues that security for highly autonomous agents "cannot be session based". When access is revoked that way, he says, the token can live on for, say, 15 minutes, 30 minutes or an hour, and "an agent can do so much damage during that time", so every agent in the ecosystem needs a kill switch. He names the shared signal framework and continuous access evaluation as technology that "has started to evolve" for this. Roy Akerman plays the attacker who gets in and out before behavioural monitoring connects the dots, and one of the hosts says SOC teams often answer with a mass credential reset because they "don't quite know what went on". Agnihotri's rapid-fire hard truth makes the same point: "the more money your agent makes, the faster you should be ready to kill it." He does not say which tokens, sessions or grants the switch would have to reach.

    An AI agent can do its damage within one session, so a practitioner wants a kill switch on its whole identityBI-FIND-2026-0134

  3. A delegated mission's authority must be a subset of the parent's, McGuinness argues, and OAuth token exchange (RFC 8693) does not guarantee it. Token exchange carries the delegation, but he calls enforcing the narrowing implementation-dependent and not a protocol guarantee. He writes that in production token exchange deployments today, child tokens routinely inherit the parent's full scope unless the implementation explicitly narrows them, so agents compound a failure that OAuth delegation already has. A child agent may also run in a different runtime or expose weaker telemetry, so authority that is narrower on paper can be less governable in practice. He proposes that the mission's state owner rate each hop's trust quality, meaning how well that hop can be observed and checked, and give a low-rated child a more conservative budget. For a sub-agent an orchestrator spawns at runtime, he says the state owner must still issue its authority, but asks what triggers that issuance. Our briefing on Uber's production design found a token chain that records each agent hop without describing any narrowing. Our briefing on Highflame's open-source ZeroID found narrowing by scope intersection offered as an explicit step.

    OAuth token exchange does not guarantee a sub-agent stays inside its parent's accessBI-FIND-2026-0228

›Cited in 29 briefings

Every briefing that discusses this term, newest first.

  1. CSA's John Yeoh expects lifecycle management to be the first agent control to break

    Podcast · · The Non-Human & AI Identity Podcast · NHI & AI Identity Summit : The Next 24 Months · Sep 7, 2026

  2. An AI agent can do its damage within one session, so a practitioner wants a kill switch on its whole identity

    Podcast · · Identity Decoded · Beyond the vault: Why AI agents force us to rethink Privileged Access Management (PAM) · Jul 15, 2026

  3. OAuth token exchange does not guarantee a sub-agent stays inside its parent's access

    Article · · Control Plane by Karl McGuinness · Karl McGuinness · Mar 18, 2026

  4. WIMSE's agent draft adds how an agent's access ends and where policy engines fit, without a new requirement

    Draft · · IETF WIMSE Working Group · Pieter Kasselman et al. · Sep 15, 2026

  5. Give an agent access for the task and take it back afterwards, four voices say, and four say people approving each request cannot keep up

    Field signals · · 8 voices from the field · Feb 2026 to Oct 2026

  6. Global Token Revocation must end a user's refresh tokens; their access tokens can outlive the call

    Draft · · IETF Datatracker · Aaron Parecki, Okta · Feb 25, 2026

  7. Receiver acknowledged, gateways cut off: in a fixed 16-host replay, 112 of 225 revoked tool calls still committed

    Paper · · arXiv · Lifei Liu et al. · Oct 11, 2026

  8. OAuth scopes grant agents far more than the task needs, four voices say, and authorization handed from agent to agent has no settled pattern yet

    Field signals · · 7 voices from the field · Oct 2025 to Oct 2026

  9. A point-in-time access review can't keep pace with changing access, four voices say, and four vendors now put AI inside the review

    Field signals · · 9 voices from the field · Nov 2024 to Aug 2026

  10. Six companies' agent-auth draft, since adopted by the WIMSE group, keeps credentials out of the model's reach

    Draft · · IETF Internet-Draft (individual submission, since adopted by the WIMSE working group) · Pieter Kasselman et al. · Jul 6, 2026

  11. Schrock's receipts prove a named key approved one exact agent action, and the draft lists at length what they cannot prove

    Draft · · IETF Datatracker · Iman Schrock, EMILIA Protocol · Sep 12, 2026

  12. The PEDIGREE draft has verifiers re-check every ancestor of an agent's token, without saying how to fetch them

    Draft · · IETF Datatracker · Karthik Rampalli, Glyphzero Labs · Apr 25, 2026

  13. NIST and CISA advise one-hour tokens, 90-day keys and hardware key storage; AI agents get one paragraph

    Article · · NIST (CSRC) · Ryan Galluzzo et al. · Sep 15, 2026

  14. Saviynt's strategy head calls agent discovery a commodity and pitches Zuma for runtime control

    Podcast · · The Non-Human & AI Identity Podcast · Ep # 14 - Turning an AI Breach into a Case Study · Sep 15, 2026

  15. A survey rates only workload identity ready for agents, and argues no standard can tell what an agent intends

    Paper · · arXiv · Takumi Otsuka et al. · Apr 25, 2026

  16. Vendors all claim to cover AI agents and non-human identities, so buyers test them in their own environment

    Podcast · · The Non-Human & AI Identity Podcast · NHI & AI Identity Summit : The Buyer's Dilemma · Sep 7, 2026

  17. Eight panels agree an agent acts for a person, never as that person. When agents hand work to other agents, nobody had an answer.

    Conference · · Las Vegas · June 2026, a one-day summit track inside Identiverse 2026; the eight panels were published as podcast episodes between 18 August and 7 September 2026

  18. A third SCIM agent draft gives the agent five lifecycle states, a quarantine, and an archive it can never leave

    Draft · · IETF Datatracker · Saurabh Kushwaha, Oracle · Jul 27, 2026

  19. Keeping an agent's delegation chain is optional, and room in the HTTP header decides whether it survives

    Podcast · · The Non-Human & AI Identity Podcast · NHI & AI Identity Summit : The New NHI Risk Reality · Aug 26, 2026

  20. Highflame's open-source ZeroID narrows an agent's token at every hop, a step we did not find in Uber's write-up

    Article · · Highflame (GitHub) · Sep 25, 2026

  21. SailPoint's product chief wants an agent's access checked before each tool call and again at each resource access

    Podcast · · The Security Strategist · Who Owns an AI Agent? Rethinking Identity, Security, & Accountability · Aug 27, 2026

  22. A valid delegation chain cannot replace required human approval or proof of key possession

    Draft · · IETF Datatracker · Karthik Rampalli, Glyphzero Labs · Jul 6, 2026

  23. A proposed standard for agent authority advises against delegating it across company lines by default

    Paper · · arXiv · Partha Madhira · May 12, 2026

  24. OAuth's cross-device guidance ranks FIDO first, then CIBA backchannel login, and the device-code flow last

    Draft · · IETF Datatracker · Pieter Kasselman et al. · Jan 23, 2026

  25. OpenID Foundation finalized Shared Signals, CAEP and RISC in September 2025, naming adopters without evidence

    Article · · OpenID Foundation · Serj Hallam · Sep 16, 2025

  26. CAEP 1.0 tells apps what changed after login and leaves almost every response to the receiver

    Article · · OpenID Foundation · Tim Cappalli et al. · Aug 29, 2025

  27. Under Shared Signals 1.0 a sender may accept a subscription and send nothing, so agent revocation needs a backup

    Article · · OpenID Foundation · Atul Tulshibagwale et al. · Aug 29, 2025

  28. Sean O'Dell says Shared Signals audiences have stopped asking what CAEP is and started asking how to deploy it

    Podcast · · Identity at the Center · #361 - Identiverse 2025 - Sean O'Dell on Harnessing CAEP Abilities with Event-Driven Identity · Jul 14, 2025

  29. Ian Glazer's sketch of a 2025 IAM stack: one shared identity data layer, with Shared Signals announcing changes

    Podcast · · Identity at the Center · #325 - Theorycrafting Modern Identity Architecture with Ian Glazer · Jan 13, 2025

In our columns

  1. Agents outrun the quarterly access review, so log every event-triggered check

    Column · Oct 28, 2026

Related terms

What appears alongside CAEP in the same briefings.

SSFRISCTransaction TokensAuthZENSaviyntSPIFFECIBAWIMSE

Was this explainer useful?

Share this explainer

Pass it on. Every claim has a permanent number, so what you share today still points at the same words next year.