]> Defakto Security

pieter@defakto.security

AWS

jeffsec@amazon.com

Zscaler

yrosomakho@zscaler.com

Ping Identity

bcampbell@pingidentity.com

OpenAI

steele@openai.com

Okta

aaron@parecki.com

WIMSE OAuth WIMSE SPIFFE Shared Signals Framework AI Agent Authentication and Authorization This document proposes best practices for authentication and authorization of AI agent interactions. It leverages existing standards such as the Workload Identity in Multi-System Environments (WIMSE) architecture and OAuth 2.0 family of specifications. Rather than defining new protocols, this document describes how existing and widely deployed standards can be applied or extended to establish agent authentication and authorization. By doing so, it aims to provide a framework within which to use existing standards, identify gaps and guide future standardization efforts for agent authentication and authorization. About This Document The latest revision of this draft can be found at . Status information for this document may be found at . Discussion of this document takes place on the WIMSE Working Group mailing list (), which is archived at . Subscribe at . Source for this draft and an issue tracker can be found at .

Introduction The rapid emergence of AI agents as autonomous workloads has sparked considerable innovation in authentication and authorization approaches. However, many of these efforts develop solutions in isolation, often reinventing existing mechanisms unaware of applicable prior art. This fragmentation risks creating incompatible implementations, duplicated development effort, and missed opportunities to leverage decades of established identity and authorization standards. This document aims to help close that gap by providing a comprehensive model demonstrating how existing, well-established standards and some emergent specifications can be composed and applied to solve agent authentication and authorization challenges. Rather than proposing new protocols, this work focuses on integrating proven standards into a coherent framework tailored to the specific requirements of AI agent workloads. By doing so, this document serves two complementary goals:

1. Consolidation of prior art: It establishes a baseline by showing how existing standards address the core identity, authentication, authorization, monitoring and observability needs of agent-based systems. Implementers and standards developers can reference this framework to avoid redundant work and ensure interoperability.
2. Foundation for future work: As the agent ecosystem matures, having such a framework aids in identifying gaps and clarifies where extensions or profiles of existing standards are needed. This provides a foundation for more focused standardization efforts in areas needing novel work rather than variations of existing approaches.

Document Status This document profiles existing standards and emerging specifications from the IETF, CNCF, and OpenID Foundation, as well as third-party protocols and mechanisms, for the AI agent use case. It defines best current practices for combining these and identifies new requirements specific to agent authentication and authorization.

Conventions and Definitions The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 when, and only when, they appear in all capitals, as shown here.

Agents are workloads An Agent is a workload that iteratively interacts with a Large Language Model (LLM) and a set of Tools, Services and Resources. An agent performs its operations until a terminating condition, determined either by the LLM or by the agent's internal logic, is reached. It may receive input from a user, or act autonomously. shows a conceptual model of the AI Agent as a workload and illustrates the high-level interaction model between the User or System, the AI Agent, the Large Language Model (LLM), Tools, Services, and Resources. In this document, Tools, Services, and Resources are treated as a single category of external endpoints that an agent invokes or interacts with to complete a task. Communication within or between Tools, Services, and Resources is out of scope.

AI Agent as a Workload | AI Agent |--(4)->| Tools | | or | | (workload) | | Services | | System |<-(6)--| |<-(5)--| Resources | +--------+ +------------+ +-----------+ ]]>

1. Optional: The User or System (e.g. a batch job or another Agent) provides an initial request or instruction to the AI Agent.
2. The AI Agent provides the available context to the LLM. Context is implementation, and deployment, specific and may include User or System input, system prompts, Tool descriptions, prior Tool, Service and Resource outputs, and other relevant state.
3. The LLM returns output to the AI Agent facilitating selection of Tools, Services or Resources to invoke.
4. The AI Agent invokes one or more external endpoints of selected Tools, Services or Resources. A Tool endpoint may itself be implemented by another AI agent.
5. The external endpoint of the Tools, Services or Resources returns a result of the operation to the AI Agent, which may send the information as additional context to the Large Language Model, repeating steps 2-5 until the exit condition is reached and the task is completed.
6. Optional: Once the exit condition is reached in step 5, the AI Agent may return a response to the User or System. The AI Agent may also return intermediate results or request additional input.

As shown in , the AI agent is a workload that needs an identifier and credentials so it can be authenticated by the Tools, Services, Resources, Large Language Model, System and the User (via the underlying operating system or platform, similar to existing applications and services). Once authenticated, these parties determine if the AI Agent is authorized to access the requested Large Language Model, Tools, Services or Resources. If the AI Agent is acting on behalf of a User or System, the User or System needs to delegate authority to the AI Agent, and the User or System context is preserved and used as input to authorization decisions and recorded in audit trails. Note: The arrows in indicate the direction in which information flows between the participants and do not imply a particular interaction style. The model applies equally to request and response exchanges and to streaming interactions. This document describes how AI Agents should leverage existing standards defined by SPIFFE , WIMSE, OAuth and OpenID SSF .

Agent Identity Management System This document defines the term Agent Identity Management System (AIMS) as a conceptual model describing the set of functions required to establish, maintain, and evaluate the identity and permissions of an agent workload. AIMS does not refer to a single product, protocol, or deployment architecture. AIMS may be implemented by one component or distributed across multiple systems (such as identity providers, provisioning services, authorization servers, policy engines, and runtime enforcement points). AIMS ensures that the right Agent has access to the right resources and tools at the right time for the right reason. Such a system depends on the following components to achieve its goals:

• Agent Identifiers: Unique identifier assigned to every Agent.
• Agent Credentials: Cryptographic binding between the Agent Identifier and attributes of the Agent.
• Agent Credential Provisioning: The mechanism for provisioning credentials to the agent at runtime.
• Agent Authentication: Protocols and mechanisms used by the Agent to authenticate itself to Large Language Models or Tools (resource or server) in the system.
• Agent Authorization: Protocols and systems used to determine if an Agent is allowed to access a Large Language Model or Tool (resource or server).
• Agent Observability and Remediation: Protocols and mechanisms to dynamically modify the authorization decisions based on observed behavior and system state.
• Agent Authentication and Authorization Policy: The configuration and rules for each of the AIMS components.
• Agent Compliance: Measurement of the state and functioning of the system against the stated policies.

The components form a logical stack in which higher layers depend on guarantees provided by lower layers, as illustrated in .

Agent Identity Management System

Agent Identifier Agents MUST be uniquely identified in order to support authentication, authorization, auditing, and delegation. The Workload Identity in Multi-System Environments (WIMSE) identifier as defined by is the primary identifier for agents in this framework. A WIMSE identifier is a URI that uniquely identifies a workload within a trust domain. Authorization decisions, delegation semantics, and audit records rely on this identifier remaining stable for the lifetime of the workload identity. The Secure Production Identity Framework for Everyone () identifier is a widely deployed and operationally mature instance of the WIMSE identifier model . A SPIFFE identifier () is a URI in the form of spiffe://<trust-domain>/<path> that uniquely identifies a workload within a trust domain. An agent participating in this framework MUST be assigned exactly one WIMSE identifier, which MAY be a SPIFFE ID. The lifecycle of an agent identifier is distinct from the lifecycle of the credentials bound to it. Credentials are short-lived and expire while the identifier persists. When an agent is retired, its identifier must not be reassigned to a different agent, and the records that reference it (audit logs, authorization decisions, delegation records and security signals) remain interpretable for as long as the deployment's retention policy requires, which is typically longer than the agent's operational lifetime. The registry or namespace in which agent identifiers are allocated, and the retention period applied to retired identifiers, are deployment-specific and out of scope for this document.

Agent Credentials Agents MUST possess credentials that provide a cryptographic binding to the agent identifier. These credentials are considered primary credentials that are provisioned at runtime. An identifier alone is insufficient unless it can be verified to be controlled by the communicating agent through a cryptographic binding. WIMSE defines two credential formats in : the Workload Identity Token (WIT) and the transport-layer X.509 Workload Identity Certificate (WIC), both of which bind a cryptographic key to a workload identifier for use in authentication. SPIFFE defines three credential formats: the X.509-SVID , WIT-SVID , and JWT-SVID . X.509-SVID is compatible with WIC, while WIT-SVID profiles WIT. JWT-SVIDs are short-lived bearer credentials used for JWT-based authentication. The choice of credential format depends on the deployment’s trust model and integration requirements. Agent credentials SHOULD be short-lived to minimize the risk of credential theft, MUST include an explicit expiration time after which it is no longer accepted, and MAY carry additional attributes relevant to the agent (for example trust domain, attestation evidence, or workload metadata). Deployments can improve the assurance of agent identity by protecting private keys using hardware-backed or isolated cryptographic storage such as TPMs, secure enclaves, or platform security modules when such capabilities are available. These mechanisms reduce key exfiltration risk but are not required for interoperability. In some cases, agents MAY need secondary credentials to access a proprietary or legacy environment that is not compatible with the X.509, JWT or WIT it is provisioned with. In these cases an agent MAY exchange their primary credentials through a credential exchange mechanisms (e.g., OAuth 2.0 Token Exchange , Transaction Tokens or Workload Identity Federation). This allows an agent to obtain a credential targeted to a specific environment by leveraging the primary credential in its possession. Note: Static API keys are an antipattern for agent identity. They are bearer artifacts that are not cryptographically bound, do not convey identity, are typically long-lived and are operationally difficult to rotate, making them unsuitable for secure agent authentication or authorization.

Agent Credential Provisioning Agent credential provisioning refers to the runtime issuance, renewal, lifecycle state and rotation of the credentials an agent uses to authenticate and authorize itself to other agents. Agents may be provisioned with one or more credential types as described in . Unlike static secrets, agent credentials are provisioned dynamically and are intentionally short-lived, eliminating the operational burden of manual expiration management and reducing the impact of credential compromise. Credential provisioning is also the assessment point to confirm that the Agent and its runtime environment satisfy the posture requirements. In this document, posture assessment refers to the evaluation of signals about the Agent, its software, deployment context, runtime environment, and operational state. These signals can influence whether a credential is issued, which identifier is bound to the credential, what type of credential is issued, what attributes are included in the credential, and how long the credential remains valid. Posture assessment mechanisms are deployment and risk specific. They may include hardware-backed evidence, trusted execution environment (TEE) evidence, software integrity measurements, supply-chain provenance, platform or orchestration-layer metadata, workload placement information, configuration state, operator assertions, or other environment-specific signals. Depending on the risk involved, a single signal may be sufficient, while higher-risk deployments may require multiple independent signals before credentials are issued or renewed. Deployed workload identity systems commonly perform some form of posture assessment as part of credential provisioning. For example, SPIFFE implementations can evaluate platform and environment-specific information about a workload before binding it to a SPIFFE identifier and issuing an SVID. At a high level, a provisioning component gathers workload and execution context signals, verifies them according to local policy, and, if verification succeeds, issues short-lived credentials for subsequent authentication and authorization. An AIMS implementation may incorporate multiple posture assessment mechanisms and implementations. The selection of mechanisms depends on deployment constraints, such as the underlying platform, available identity signals, and desired level of trust assurance. This document does not require any particular posture assessment mechanism, evidence format, or verifier architecture. Agent credential provisioning must operate autonomously, scale to high-churn environments, and integrate closely with the posture assessment mechanisms used to establish trust in the Agent at each issuance or rotation event. Posture assessment at provisioning time establishes trust at the moment a credential is issued or renewed. It does not by itself establish that the posture still holds when a particular request is made. Higher-risk deployments, and interactions that cross trust domain boundaries, may require posture evidence that is closer to the time of the request. Two approaches are available with existing mechanisms:

1. Shortening credential lifetimes so that renewal, and with it re-assessment, occurs more frequently.
2. Using the eventing mechanisms described in so that a change in posture detected after issuance results in revocation or down-scoping of the agent's credentials and tokens.

Agent credential provisioning typically includes two phases:

1. Initial Provisioning: The process by which an agent first acquires a credential bound to its identity. This often occurs immediately after deployment or instantiation and is based on verified properties of the agent (e.g., deployment context, posture assessment results, or orchestration metadata).
2. Rotation/Renewal: The automatic refresh of short-lived credentials before expiration. Continuous rotation ensures that credentials remain valid only for the minimum necessary time and that authorization state reflects current operational conditions.
3. Deprovisioning: The termination of an agent's credentials when the agent is retired, when its posture no longer satisfies policy, or when its identity is believed to be compromised. At a minimum, deprovisioning stops renewal so that outstanding short-lived credentials expire and are not replaced. Where the change must take effect before expiry, the issuer signals the status change to relying parties and to the issuers of any credentials derived from the agent's primary credential, so that the derived credentials are revoked or cease to be accepted as well.

The use of short-lived credentials provides a significant improvement in the risk profile and risk of credential exposure. It provides an alternative to explicit revocation mechanisms and simplifies lifecycle management in large, automated environments while removing the risks of downtime as a result of credential expiry. Deployed frameworks such as provide proven mechanisms for automated, short-lived credential provisioning at runtime. In addition to issuing short-lived credentials, also provisions ephemeral cryptographic key material bound to each credential, further reducing the risks associated with compromising long-lived keys. The Large Language Model MUST NOT have access to an agent's credentials or to credentials that may be needed to access tools and services. This prevents the Large Language Model from using, exposing, or being manipulated via prompt injection into disclosing the credentials.

Agent Authentication Agents may authenticate using a variety of mechanisms, depending on the credentials they possess, the protocols supported in the deployment environment, and the risk profile of the application. As described in the WIMSE Architecture , authentication can occur at either the transport layer or the application layer, and many deployments rely on a combination of both.

Transport Layer Authentication Transport-layer authentication establishes trust during the establishment of a secure transport channel. The most common mechanism used by agents is mutually-authenticated TLS (mTLS), in which both endpoints present X.509-based credentials and perform a bidirectional certificate exchange as part of the TLS negotiation. When paired with short-lived workload identities, such as those issued by SPIFFE or WIMSE, mTLS provides strong channel binding and cryptographic proof of control over the agent’s private key. mTLS is particularly well-suited for environments where transport-level protection, peer authentication, and ephemeral workload identity are jointly required. It also simplifies authorization decisions by enabling agents to associate application-layer requests with an authenticated transport identity. One example of this is the use of mTLS in service mesh architectures such as Istio or LinkerD.

Limitations There are scenarios where transport-layer authentication is not desirable or cannot be relied upon. In architectures involving intermediaries, such as proxies, API gateways, service meshes, load balancers, or protocol translators, TLS sessions are often terminated and re-established, breaking the end-to-end continuity of transport-layer identity. Similarly, some deployment models (such as serverless platforms, multi-tenant edge environments, or cross-domain topologies) may obscure or abstract identity presented at the transport layer, making it difficult to bind application-layer actions to a credential presented at the transport layer. In these cases, application-layer authentication provides a more robust and portable mechanism for expressing agent identity and conveying attestation or policy-relevant attributes.

Application Layer Authentication Application-layer authentication allows agents to authenticate independently of the underlying transport. This enables end-to-end identity preservation even when requests traverse proxies, load balancers, or protocol translation layers. The WIMSE working group defines WIMSE Proof Tokens and HTTP Message Signatures as authentication mechanisms that may be used by agents.

WIMSE Proof Tokens (WPTs) WIMSE Workload Proof Tokens (WPTs, ) are a protocol-independent, application-layer mechanism for proving possession of the private key associated with a Workload Identity Token (WIT). WPTs are generated by the agent, using the private key matching the public key in the WIT. A WPT is defined as a signed JSON Web Token (JWT) that binds an agent’s authentication to a specific message context, for example, an HTTP request, thereby providing proof of possession rather than relying on bearer semantics. WPTs are designed to work alongside WITs and are typically short-lived to reduce the window for replay attacks. They carry claims such as audience (aud), expiration (exp), a unique token identifier (jti), and a hash of the associated WIT (wth). A WPT may also include hashes of other related tokens (e.g., a Transaction Token) to bind the authentication contexts to specific transaction or authorizations details. Although the draft currently defines a protocol binding for HTTP (via a Workload-Proof-Token header), the core format is protocol-agnostic, making it applicable to other protocols. Its JWT structure and claims model allow WPTs to be bound to different protocols and transports, including asynchronous or non-HTTP messaging systems such as Kafka and gRPC, or other future protocol bindings. This design enables receiving systems to verify identity, key possession, and message binding at the application layer even in environments where transport-layer identity (e.g., mutual TLS) is insufficient or unavailable.

HTTP Message Signatures The WIMSE Workload-to-Workload Authentication with HTTP Signatures specification defines an application-layer authentication profile built on the HTTP Message Signatures standard . It is one of the mechanisms WIMSE defines for authenticating workloads in HTTP-based interactions where transport-layer protections may be insufficient or unavailable. The protocol combines a workload's Workload Identity Token (WIT) (which binds the agent's identity to a public key) with HTTP Message Signatures (using the corresponding private key), thereby providing proof of possession and message integrity for individual HTTP requests and responses. This approach ensures end-to-end authentication and integrity even when traffic traverses intermediaries such as TLS proxies or load balancers that break transport-layer identity continuity. The profile mandates signing of some request components (e.g., method, request-target, content digest, and the WIT itself) and supports optional response signing. Note that @request-target covers only the request-target string (typically path + query) and not the method, scheme, or authority; those are only protected if separately covered (e.g., @method, @scheme, @authority).

Limitations Unlike transport-layer authentication, application-layer authentication does not inherently provide channel binding to the underlying secure transport. As a result, implementations MUST consider the risk of message relay or replay if tokens or signed messages are accepted outside their intended context. Deployments typically mitigate these risks through short token lifetimes, audience restrictions, nonce or unique identifier checks, and binding authentication to specific requests or transaction parameters.

Long-Lived Sessions and Streams In a long lived session or a streaming connection, the Agent is authenticated when the session is established, using the mechanisms described above, and the authorization granted at that point is bounded by the lifetime of the credential or token presented. Because a session may outlive the credential or token used to establish it, the Agent keeps its credentials current through the automated rotation described in , the party accepting the session re-validates authorization when the credential or token expires, and both act on revocation or risk signals () by terminating or down-scoping the session rather than allowing it to run to completion. Where individual messages within a session require their own proof of possession, WPTs () can be applied to those messages.

Agent Authorization Agents act on behalf of a user, a system, or on their own behalf as shown in and need to obtain authorization when interacting with protected resources.

Agent Mission An Agent receives a Mission from a User, a System, or another Agent. The Mission is the task or objective the Agent will pursue. It is typically expressed in natural language and may require decomposition as it is mapped from a broad objective into specific resource requirements and corresponding access requests. The translation of a mission into authorization requirements is expected to occur as a planning step before the Agent requests authorization. The process through which the mission is translated into authorization requriements is out of scope of this specification. Once the required resources and their associated authorization requirements are determined, the Agent SHOULD use the mechanisms described in this section to obtain access.

Leverage OAuth 2.0 as a Delegation Authorization Framework The widely deployed OAuth 2.0 Authorization Framework is a mechanism for delegated authorization that enables an Agent to obtain limited access to a protected resource (e.g., a service or API), intermediated by an Authorization Server, often with the explicit approval of the authenticated User. An Agent uses OAuth 2.0-based mechanisms to obtain authorization from a User, a System, or on its own behalf. OAuth 2.0 defines a wide range of authorization grant flows that supports these scenarios. In these Oauth 2.0 flows, an Agent acts as an OAuth 2.0 Client to an OAuth 2.0 Authorization Server, which receives the request, evaluate the authorization policy and returns an access token, which the Agent presents to the Resource Server (i.e. the protected resources such as the LLM or Tools in , which can evaluate its authorization policy and complete the request.

Use of OAuth 2.0 Access Tokens An OAuth access token represents the authorization granted to the Agent. In many deployments, access tokens are structured as JSON Web Tokens (JWTs) , which include claims such as 'client_id', 'sub', 'aud', 'scope', and other attributes relevant to authorization. The access token includes the Agent identity as the 'client_id' claim as defined in . When the Agent is acting on-behalf of another User or System, the User or System identifier is conveyed in the 'sub' claim as defined in . These identifiers MUST be used by resource servers protected by the OAuth 2.0 authorization service, along with other claims in the access token, to determine if access to a resource should be allowed. The access token typically includes additional claims to convey contextual, attestation-derived, or policy-related information that enables fine-grained access control. The resource server uses the access token and the information it contains along with other inputs, as described in , when enforcing access control. JWT access tokens can be validated directly by resource servers while other formats that are opaque to the resource server can be validated through a mechanism that calls back to the authorization server (the mechanism is called introspection despite the word having nearly the opposite meaning). This framework supports both models and does not require a specific token format, provided that equivalent authorization semantics are maintained. A resource server in receipt of tokens opaque to it are able to obtain authorization and other information from the token through OAuth 2.0 Token Introspection . The introspection response provides the active state of the token and associated authorization attributes equivalent to those conveyed in structured tokens.

Interaction with Policy-Based Systems In some deployments the claims in an access token are sufficient for a resource server to make an authorization decision on their own. Where they are not, the access token becomes one input to a decision made by a policy-based authorization system (for example a policy-, attribute- or role-based system), which evaluates it together with other inputs. The authenticated identity of the requester (), the resource and action requested, the delegated User or System context, posture and risk signals (, ), and environmental attributes such as time or network location. The policy itself, and the language in which it is expressed, are local to the resource server and its trust domain. Within a trust domain the components are operated by one organization and the use of a policy langugage within that domain is a deployment choice. Across trust domains, the mechanisms in convey the outcome of a policy decision made in one domain to another domain as a token with defined claims and scope, rather than conveying the policy itself. Each domain continues to evaluate its own policy against the tokens it receives. A common policy language is therefore not required for the mechanisms in this document to interoperate across trust domains. Where policy is managed centrally or shared by several resource servers, a resource server may request decisions from an external policy decision point rather than evaluating policy itself. A common interface for that interaction is defined by the OpenID AuthZEN Authorization API . Access tokens may be sender-constrained to mitigate the risk of accidental token leakage or deliberate token theft, so that a token obtained by an attacker cannot be used without the corresponding key. Sender-constraining may use certificate-bound access tokens , when the Agent presents an X.509 credential over mutual TLS, or Demonstrating Proof of Possession (DPoP) , when transport-layer binding is not feasible. The use of sender-constrained access tokens is described in the OAuth 2.0 Security Best Current Practice and the FAPI 2.0 Security Profile .

Obtaining an OAuth 2.0 Access Token OAuth 2.0 defines a number authorization grant flows in support of different authorization scenarios. The appropriate flow depends on the specific authorization scenario and the nature of User involvement. The following subsections describe the most relevant flows for Agent authorization.

User Delegates Authorization When a User delegates authorization to an Agent, the Authorization Code Grant as described in Section 4.1 of is the appropriate mechanism. This redirection-based flow involves an interactive authorization process, typically in a web browser, in which the user authenticates to the authorization server and explicitly approves the requested access. Users SHOULD be authenticated using phishing-resistant authentication mechanisms such as a passkey. Once the user has approved the request, the authorization server returns an authorization code to the Agent via the redirect. The Agent, acting as an OAuth client, then makes a token request to the authorization server to redeem the authorization code for an access token. When making this token request, the Agent authenticates itself directly to the authorization server using the credentials described in with a compatible OAuth client authentication mechanism, and not with the use of static, long-lived client secrets. Compatible OAuth client authentication mechanisms are defined in , and . The OAuth client authentication step is distinct from, and occurs after, the user authentication and approval described above. The resulting access token reflects the authorization delegated to the Agent by the User and can be used by the Agent to access resources on behalf of the user. The use of OAuth negates the need for the Agent to have access to a User's credentials when accessing a resource on the User's behalf.

Agent Obtains Own Authorization Agents obtaining access tokens on their own behalf can use the Client Credentials Grant as described in or the JWT Authorization Grant as described in . When using the Client Credentials Grant, the Agent authenticates itself using the credentials described in with a compatible OAuth client authentication mechanism listed in previous section, and not with the use of static, long-lived client secrets. When using the JWT Authorization Grant, the Agent will be identified in the subject of the JWT assertion.

Agents Accessed by Systems or Other Agents Agents themselves can act in the role of an OAuth protected resource and be invoked by a System (e.g. a batch job or another Agent). The System obtains an access token using an appropriate mechanism and then invokes the Agent presenting the access token.

OAuth 2.0 Security Best Practices The Best Current Practice for OAuth 2.0 Security as described in are applicable when requesting and using access tokens.

Risk Reduction with Transaction Tokens Resources servers, whether they are LLMs, Tools or Agents (in the Agent-to-Agent case) may be composed of multiple microservices that are invoked to complete a request. The access tokens presented to the Agent, LLM or Tools can typically be used with multiple transactions and consequently have broader scope than needed to complete any specific transaction. Passing the access token from one microservice to another within an invoked Agent, LLM or the Tools increases the risk of token theft and replay attacks. For example, an attacker may discover and access token passed between microservices in a log file or crash dump, exfiltrate it, and use it to invoke a new transaction with different parameters (e.g. increase the transaction amount, or invoke an unrelated call as part of executing a lateral move). To avoid passing access tokens between microservices, the Agent, LLM or Tools can exchange the received access token for a transaction token, as defined in the Transaction Token specification . The transaction token allows for identity and authorization information to be passed along the internal call chain of microservices. The transaction token issuer enriches the transaction token with context of the caller that presented the access token (e.g. IP address, etc.), transaction context (transaction amount), identity information and a unique transaction identifier. This results in a downscoped token that is bound to a specific transaction and cannot be used as an access token, with another transaction, or within the same transaction with modified transaction details (e.g. change in transaction amount). Transaction tokens are typically short-lived, further limiting the risk in case they are obtained by an attacker by limiting the time window during which these tokens will be accepted. The scope of access conveyed by each token in this sequence of exchanges can never exceed that of the token presented to obtain it. At each exchange the authorization server (or transaction token service) evaluates its authorization policy and makes the final decision on whether to issue a token and with what scope. A transaction token may be presented in an OAuth 2.0 Token Exchange to obtain an access token for another service (e.g. another Agent, Tool or LLM). Obtaining an access token for a service in another trust domain when the Agent has a transaction is described in .

Cross Domain Access Agents often require access to resources that are protected by different OAuth 2.0 authorization servers. When the components in Figure 1 are protected by different logical authorization servers, an Agent SHOULD use OAuth Identity and Authorization Chaining Across Domains as defined in , or a derived specification such as the Identity Assertion JWT Authorization Grant or , Transaction Token Authorization Grant Profile for OAuth Identity and Authorization Chaining , to obtain an access token from the relevant authorization servers. The agent first exchanges its current access token or transaction token for a JWT authorization grant (Section 2.3 of ) and then presents that grant to obtain an access token for the target resource (Section 2.4 of ). When using the Identity Assertion JWT Authorization Grant , the identity assertion (e.g. the OpenID Connect ID Token or SAML assertion) for the target end-user is used to obtain a JWT assertion as described in , which is then used to obtain an access token as described in . OAuth Identity and Authorization Chaining Across Domains () provides a general mechanism for obtaining cross-domain access that can be used whether an identity assertion like a SAML or OpenID Connect token is available or not. The Identity Assertion JWT Authorization Grant is optimized for cases where an identity assertion like a SAML or OpenID Connect token is available from an identity provider that is trusted by all the OAuth authorization servers as it removes the need for the user to re-authenticate. This is typically used within enterprise deployments to simplify authorization delegation for multiple software-as-a-service offerings.

Human in the Loop An OAuth authorization server MAY conclude that the level of access requested by an Agent requires explicit user confirmation. In such cases the authorization server SHOULD either decline the request or obtain additional authorization from the User. An Agent, acting as an OAuth client, can use the OpenID Client Initiated Backchannel Authentication (CIBA) protocol (). This triggers an out-of-band interaction allowing the user to approve or deny the requested operation without exposing credentials to the agent (for example a push notification requesting the user to approve a request through an authenticator application on their mobile device). Interactive agent frameworks may also solicit user confirmation directly during task execution (for example tool invocation approval or parameter confirmation). Such interactions do not by themselves constitute authorization and MUST be bound to a verifiable authorization grant issued by the authorization server. The agent SHOULD therefore translate user confirmation into an OAuth authorization event (e.g., step-up authorization via CIBA) before accessing protected resources. This model aligns with user-solicitation patterns such as those described by the Model Context Protocol (), where an agent pauses execution and requests user confirmation before performing sensitive actions. The final authorization decision remains with the authorization server, and the agent MUST NOT treat local UI confirmation alone as sufficient authorization. Note: Additional specification or design work may be needed to define how out-of-band interactions with the User occur at different stages of execution. CIBA itself only accounts for client initiation, which doesn't map well to cases that envision the need for User confirmation to occur mid-execution.

Tool-to-Service Access Tools expose interfaces to underlying services and resources. Access to the Tools can be controlled by OAuth and augmented by policy, attribute or role based authorization systems (amongst others). If the Tools are implemented as one or more microservices, it should use transaction tokens to reduce risk as described in to avoid passing access tokens around within the Tool implementation. Access from the Tools to the resources and services MAY be controlled through a variety of authorization mechanisms, including OAuth. If access is controlled through OAuth, the Tools can use OAuth 2.0 Token Exchange as defined in to exchange the access token it received for a new access token to access the resource or service in question. When the Tool needs access to a resource protected by an authorization server other than the Tool's own authorization server, the OAuth Identity and Authorization Chaining Across Domains () can be used to obtain an access token from the authorization server protecting that resource. Note: It is an anti-pattern for Tools to forward access tokens it received from the Agent to Services or Resources. It increases the risk of credential theft and lateral attacks.

Privacy Considerations Authorization tokens may contain user identifiers, agent identifiers, audience restrictions, transaction details, and contextual attributes. Deployments SHOULD minimize disclosure of personally identifiable or sensitive information in tokens and prefer audience-restricted and short-lived tokens. Where possible, opaque tokens with introspection SHOULD be preferred when claim minimization is required. Agents SHOULD request only the minimum scopes and authorization details necessary to complete a task. As recommended in , resource servers avoid logging full tokens and instead log token identifiers or hashes. When authorization context is propagated across services, derived or down-scoped tokens (such as transaction tokens) SHOULD be used to reduce correlation and replay risk. Implementations MUST ensure that user identity information delegated to agents is not exposed to unrelated services and that cross-domain authorization exchanges only disclose information required for the target authorization decision.

OAuth 2.0 Discovery in Dynamic Environments In dynamic Agent deployments (e.g., ephemeral workloads, multi-tenant services, and frequently changing endpoint topology), Agents and other participants MAY use OAuth discovery mechanisms to reduce static configuration and to bind runtime decisions to verifiable metadata.

Authorization Server Capability Discovery An Agent that needs to obtain tokens can discover authorization server endpoints and capabilities using OAuth 2.0 Authorization Server Metadata and/or OpenID Connect Discovery . This allows the Agent to learn the authorization server issuer identifier, authorization and token endpoints, supported grant types, client authentication methods, signing keys (via jwks_uri), and other relevant capabilities without preconfiguring them.

Protected Resource Capability Discovery When an Agent is invoking a Tool, the Agent MAY use OAuth 2.0 Protected Resource Metadata to discover how the resource is protected, including the resource identifier and the applicable Authorization Server(s) that protects Tool access. This enables an Agent to select the correct issuer/audience and token acquisition flow at runtime, even when resources are deployed or moved dynamically. A Tool that attempts to access and OAuth protected resource MAY use OAuth 2.0 Protected Resource Metadata in a similar way as an Agent. Similarly, a System may use when accessing an Agent.

Client Capability Discovery Other actors (e.g., Authorization Servers, registrars, or policy systems) may need to learn about any entities (System, Agent, Tool) that acts as OAuth clients. Where supported, they MAY use Client ID Metadata Documents , which allow a client to host its metadata at a URL-valued client_id so that the relying party can retrieve client properties (e.g., redirect URIs, display information, and other registered client metadata) without prior bilateral registration. As an alternative, entities acting as OAuth clients MAY register their capabilities with authorization servers as defined in the OAuth 2.0 Dynamic Client Registration Protocol .

Agent Monitoring, Observability and Remediation Because agents may perform sensitive actions autonomously or on behalf of users, deployments MUST maintain sufficient monitoring and observability to reconstruct agent behavior and authorization context after execution. Observability is therefore a security control, not solely an operational feature. Any participant in the system, including the Agent, Tool, System, LLM or other resources and service MAY subscribe to change notifications using eventing mechanisms such as the OpenID Shared Signals Framework and its profiles to receive security and authorization-relevant signals including but not limited to Continuous Access Evaluation Profile , Risk Incident Sharing and Coordination or Workload Identity Security Events . Upon receipt of a relevant signal (e.g., session revoked, risk level change, subject disabled, token replay suspected, risk elevated), the recipient SHOULD remediate by attenuating access, such as terminating local sessions, discarding cached tokens, re-acquiring tokens with updated constraints, reducing privileges, or re-running policy evaluation before continuing to allow access. Recipients of such signals MUST ensure that revoked or downgraded authorization is enforced without undue delay. Cached authorization decisions and tokens that are no longer valid MUST NOT continue to be used after a revocation or risk notification is received. To support detection, investigation, and accountability, deployments MUST produce durable audit logs covering authorization decisions and subsequent remediations. Audit records MUST be tamper-evident and retained according to the security policy of the deployment. At a minimum, audit events MUST record:

• authenticated agent identifier
• delegated subject (user or system), when present
• resource or tool being accessed
• action requested and authorization decision
• timestamp and transaction or request correlation identifier
• posture assessment or risk state influencing the decision
• remediation or revocation events and their cause

Monitoring / Observability systems SHOULD correlate events across Agents, Tools, Services, Resources and LLMs to detect misuse patterns such as replay, confused deputy behavior, privilege escalation, or unexpected action sequences. End-to-end audit is enabled when Agents, Users, Systems, LLMs, Tools, services and resources have stable, verifiable identifiers that allow auditors to trace "which entity did what, using which authorization context, and why access changed over time." Implementations SHOULD provide operators the ability to reconstruct a complete execution chain of an agent task, including delegated authority, intermediate calls, and resulting actions across service boundaries.

Agent Authentication and Authorization Policy The configuration and runtime parameters for Agent Identifiers , Agent Credentials , Agent Credential Provisioning , Agent Authentication , Agent Authorization and Agent Monitoring, Observability and Remediation collectively constitute the authentication and authorization policy within which the Agent operates within an AIMS implementation. Because these parameters are highly deployment and risk-model-specific (and often reflect local governance, regulatory, and operational constraints), the policy model and document format are out of scope for this framework and are not recommended as a target for standardization within this specification. Implementations MAY represent policy in any suitable “policy-as-code” or configuration format (e.g., JSON/YAML), provided it is versioned, reviewable, and supports consistent evaluation across the components participating in the end-to-end flow.

Agent Compliance Compliance for Agent-based systems SHOULD be assessed by auditing observed behavior and recorded evidence (logs, signals, and authorization decisions) against the deployment’s Agent Authentication and Authorization Policy . Since compliance criteria are specific to individual deployments, organizations, industries and jurisdictions, they are out of scope for this framework though implementers SHOULD ensure strong observability and accountable governance, subject to their specific business needs.

Security Considerations This document composes existing specifications to enable authentication and authorization between AI agents. The security considerations of each referenced specification apply in full. In addition to the guidance included in existing specifications, additional security best practices and profiles have been developed for the Oauth protocol famility. The OAuth 2.0 Security Best Current Practice which captures current guidance on threats and mitigations that have emerged since the original OAuth 2.0 specifications were published. The FAPI 2.0 Security Profile defines a high-assurance profile of OAuth 2.0 suitable for high security applications.

Privacy Considerations This document composes existing specifications to enable authentication and authorization between AI agents. In addition to the the privacy considerations in , the privacy considerations in each referenced specification apply in full.

IANA Considerations This document has no IANA actions.

Acknowledgments The authors would like to thank:

• Sean O'Dell for providing valuable input and feedback on this work.
• Karl McGuinness for his blog posts on mission shaping as a pre-cursor to authorization

References Normative References n.d. n.d. Asserto Axiomatics SGNL n.d. n.d. n.d. n.d. n.d. n.d. n.d. n.d. n.d. n.d. n.d. n.d. n.d. n.d. In many standards track documents several words are used to signify the requirements in the specification. These words are often capitalized. This document defines these words as they should be interpreted in IETF documents. This document specifies an Internet Best Current Practices for the Internet Community, and requests discussion and suggestions for improvements. RFC 2119 specifies common key words that may be used in protocol specifications. This document aims to reduce the ambiguity by clarifying that only UPPERCASE usage of the key words have the defined special meanings. Zscaler Palo Alto Networks This document defines a canonical identifier for workloads, referred to as the Workload Identifier. A Workload Identifier is a URI that uniquely identifies a workload within the context of a specific trust domain. This identifier can be embedded in Workload Identity Credentials, including X.509 certificates and JWT-based tokens, to support authentication, authorization, and policy enforcement across diverse systems. The Workload Identifier format ensures interoperability, facilitates secure identity federation, and enables consistent identity semantics. Ping Identity Palo Alto Networks Defakto Security Independent Zscaler The WIMSE architecture defines authentication and authorization for software workloads in a variety of runtime environments, from the most basic ones up to complex multi-service, multi-cloud, multi- tenant deployments. This document defines the credentials that workloads use to represent their identity. They can be used in various protocols to authenticate workloads to each other. To use these credentials, workloads must provide proof of possession of the associated private key material, which is covered in other documents. This document focuses on the credentials alone, independent of the proof-of- possession mechanism. This specification defines a protocol for an HTTP- and JSON-based Security Token Service (STS) by defining how to request and obtain security tokens from OAuth 2.0 authorization servers, including security tokens employing impersonation and delegation. CrowdStrike Practical Identity LLC Defakto Security Transaction Tokens (Txn-Tokens) are designed to maintain and propagate user identity, workload identity and authorization context throughout the Call Chain within a trusted domain during the processing of external requests (e.g. such as API calls) or requests initiated internally within the Trust Domain. Txn-Tokens ensure that this context is preserved throughout the Call Chain thereby enhancing security and consistency in complex, multi-service architectures. Palo Alto Networks Zscaler University of the Bundeswehr Munich The increasing prevalence of cloud computing and micro service architectures has led to the rise of complex software functions being built and deployed as workloads, where a workload is defined as software executing for a specific purpose, potentially comprising one or more running instances. This document discusses an architecture for designing and standardizing protocols and payloads for conveying workload identity and security context information. Ping Identity Defakto Security The WIMSE architecture defines authentication and authorization for software workloads in a variety of runtime environments, from basic deployments to complex multi-service, multi-cloud, multi-tenant systems. This document specifies the Workload Proof Token (WPT), a mechanism for workloads to prove possession of the private key associated with a Workload Identity Token (WIT). The WPT is a signed JWT that binds the workload's authentication to a specific HTTP request, providing application-layer proof of possession for workload-to-workload communication. This specification is designed to work alongside the WIT credential format defined in draft-ietf- wimse-workload-creds and can be combined with other WIMSE protocols in multi-hop call chains. Palo Alto Networks Independent The WIMSE architecture defines authentication and authorization for software workloads in a variety of runtime environments, from the most basic ones to complex multi-service, multi-cloud, multi-tenant deployments. This document defines one of the mechanisms to provide workload authentication, using HTTP Signatures. While only applicable to HTTP traffic, the protocol provides end-to-end protection of requests (and optionally, responses), even when service traffic is not end-to-end encrypted, that is, when TLS proxies and load balancers are used. Authentication is based on the Workload Identity Token (WIT). This document describes a mechanism for creating, encoding, and verifying digital signatures or message authentication codes over components of an HTTP message. This mechanism supports use cases where the full HTTP message may not be known to the signer and where the message may be transformed (e.g., by intermediaries) before reaching the verifier. This document also describes a means for requesting that a signature be applied to a subsequent HTTP message in an ongoing HTTP exchange. The OAuth 2.0 authorization framework enables a third-party application to obtain limited access to an HTTP service, either on behalf of a resource owner by orchestrating an approval interaction between the resource owner and the HTTP service, or by allowing the third-party application to obtain access on its own behalf. This specification replaces and obsoletes the OAuth 1.0 protocol described in RFC 5849. [STANDARDS-TRACK] This specification defines a profile for issuing OAuth 2.0 access tokens in JSON Web Token (JWT) format. Authorization servers and resource servers from different vendors can leverage this profile to issue and consume access tokens in an interoperable manner. This specification defines a method for a protected resource to query an OAuth 2.0 authorization server to determine the active state of an OAuth 2.0 token and to determine meta-information about this token. OAuth 2.0 deployments can use this method to convey information about the authorization context of the token from the authorization server to the protected resource. This document describes a mechanism for sender-constraining OAuth 2.0 tokens via a proof-of-possession mechanism on the application level. This mechanism allows for the detection of replay attacks with access and refresh tokens. This specification defines the use of a JSON Web Token (JWT) Bearer Token as a means for requesting an OAuth 2.0 access token as well as for client authentication. This document describes OAuth client authentication and certificate-bound access and refresh tokens using mutual Transport Layer Security (TLS) authentication with X.509 certificates. OAuth clients are provided a mechanism for authentication to the authorization server using mutual TLS, based on either self-signed certificates or public key infrastructure (PKI). OAuth authorization servers are provided a mechanism for binding access tokens to a client's mutual-TLS certificate, and OAuth protected resources are provided a method for ensuring that such an access token presented to it was issued to the client presenting the token. Defakto Security Defakto Security NIST IBM Cisco Systems This specification profiles the Assertion Framework for OAuth 2.0 Client Authentication and Authorization Grants [RFC7521], the JWT Profile for OAuth 2.0 Client Authentication and Authorization Grants [RFC7523], and OAuth 2.0 Attestation-Based Client Authentication [I-D.draft-ietf-oauth-attestation-based-client-auth] to enable the use of SPIFFE Verifiable Identity Documents (SVIDs) as client credentials in OAuth 2.0. It defines how OAuth clients with SPIFFE credentials can authenticate to OAuth authorization servers using their JWT-SVIDs, WIT-SVIDs, or X509-SVIDs without the need for client secrets. This approach enhances security by enabling seamless integration between SPIFFE-enabled workloads and OAuth authorization servers while eliminating the need to distribute and manage shared secrets such as static client secrets. This document describes best current security practice for OAuth 2.0. It updates and extends the threat model and security advice given in RFCs 6749, 6750, and 6819 to incorporate practical experiences gathered since OAuth 2.0 was published and covers new threats relevant due to the broader application of OAuth 2.0. Further, it deprecates some modes of operation that are deemed less secure or even insecure. Defakto Security Defakto Security MITRE NSA-CCSS Ping Identity Okta This specification describes a mechanism for preserving identity and authorization information across trust domains that use the OAuth 2.0 Framework. A JSON Web Token (JWT) authorization grant, obtained through an intra-domain OAuth 2.0 Token Exchange, facilitates the cross-domain acquisition of an access token. The relevant identity and authorization information is chained throughout the flow by being conveyed in the respective artifacts exchanged at each step of the process. Chaining across multiple domains is achieved by using the same protocol every time a trust domain boundary is crossed. Okta Independent Ping Identity This specification provides a mechanism for an application to use an identity assertion to obtain an access token for a third-party API by coordinating through an identity provider that the downstream Resource Authorization Server already trusts for single sign-on (SSO), using OAuth 2.0 Token Exchange (RFC 8693) and the JWT Profile for OAuth 2.0 Authorization Grants (RFC 7523). This pattern is informally referred to as Cross-App Access (XAA). Practical Identity LLC Defakto Security CVS Health This specification defines a profile of the OAuth Identity and Authorization Chaining Across Domains [I-D.ietf-oauth-identity-chaining] mechanism that uses a Transaction Token (Txn-Token) [I-D.ietf-oauth-transaction-tokens] as the subject token in a Token Exchange [RFC8693] request to obtain a JWT Authorization Grant for crossing a trust boundary. A Txn-Token is scoped to a single trust domain and represents the full authorization context of an in-progress transaction, regardless of whether that transaction was initiated by a human user calling an external API, by an internal system event, or by an automated workload. This profile specifies how a service operating within that trust domain can present its Txn-Token to obtain a JWT Authorization Grant that carries the necessary context across a trust boundary, enabling an access token to be issued for a partner service, without exposing internal trust-domain credentials or token formats beyond the trust boundary. This specification defines a metadata format that an OAuth 2.0 client can use to obtain the information needed to interact with an OAuth 2.0 authorization server, including its endpoint locations and authorization server capabilities. This specification defines a metadata format that an OAuth 2.0 client or authorization server can use to obtain the information needed to interact with an OAuth 2.0 protected resource. Okta This specification defines a mechanism through which an OAuth client can identify itself to authorization servers, without prior dynamic client registration or other existing registration. This is through the usage of a URL as a client_id in an OAuth flow, where the URL refers to a document containing the necessary client metadata, enabling the authorization server to fetch the metadata about the client as needed. This specification defines mechanisms for dynamically registering OAuth 2.0 clients with authorization servers. Registration requests send a set of desired client metadata values to the authorization server. The resulting registration responses return a client identifier to use at the authorization server and the client metadata values registered for the client. The client can then use this registration information to communicate with the authorization server using the OAuth 2.0 protocol. This specification also defines a set of common client metadata fields and values for clients to use during registration. Informative References

Document History [[ To be removed from the final specification ]] -01

• Editorial updates
• Clarify relationship between WIMSE and SPIFFE credentials (https://github.com/PieterKas/agent2agent-auth-framework/issues/136)
• Clarify that LLMs should not have access to Agent credentials: https://github.com/PieterKas/agent2agent-auth-framework/issues/127
• Update use of normative language (https://github.com/PieterKas/agent2agent-auth-framework/issues/123)
• Added reference to newly adopted WISE draft (https://github.com/ietf-wg-wimse/draft-ietf-wimse-aims/issues/153)
• Address concerns about circularity of using a transaction token to get an access token (https://github.com/ietf-wg-wimse/draft-ietf-wimse-aims/issues/126)
• Explicit guidance on policy langugages and policy based systems (https://github.com/ietf-wg-wimse/draft-ietf-wimse-aims/issues/124)
• Added text on sender constraining tokens (https://github.com/ietf-wg-wimse/draft-ietf-wimse-aims/issues/126)
• Make it explicit that the AIMS draft applies regardless of streaming or request/response interactions (https://github.com/ietf-wg-wimse/draft-ietf-wimse-aims/issues/129)
• Added text to clarify that posture assessment is ongoing with forward pointing references. (https://github.com/ietf-wg-wimse/draft-ietf-wimse-aims/issues/128)
• Make deprovisioning explicit (https://github.com/ietf-wg-wimse/draft-ietf-wimse-aims/issues/125)
• Re-use the AIMS term (https://github.com/ietf-wg-wimse/draft-ietf-wimse-aims/issues/135)
• fix up some reference names

-03

• Include reference to CIBA (see https://github.com/PieterKas/agent2agent-auth-framework/issues/131)

-02

• Add Aaron Parecki from Okta as co-author.
• Add reference to phishing resistent credentials (e.g. FIDO passkeys/authenticators) - see https://github.com/PieterKas/agent2agent-auth-framework/issues/106
• Fold attestation section into provosioning section. Change terminology to posture management.
• Clarify Oauth authentication mechanisms when an Agent acts as and OAuth client.
• Update Security Considerations section
• Add section on Agent Mission (see issue https://github.com/PieterKas/agent2agent-auth-framework/issues/107)

-01

• Add Nick Steele from OpenAI as co-author.

-00