Protocol Specification
An open, cryptographically verifiable, and extensible evidence protocol for AI agents
Overview
AEP (Attestation & Evidence Exchange Protocol) is an open evidence infrastructure protocol built specifically for AI agents.
The core problem it solves: in an autonomous agent era, how to prove which agent executed a critical action, when it happened, and whether the context and execution remain intact and untampered.
Through digital certificates, causal time proofs, and global ledger anchoring, AEP establishes a transparent and neutrally auditable evidence chain for independent verification across the web.
Design Principles
Openness
An open protocol and specification; any party can freely build standards-compliant implementations
Cryptographically Verifiable
Built on proven cryptography; all claims are independently verifiable without relying on trusted intermediaries
Extensibility
Modular layered design supporting custom signature schemes and verifiable data structures
Interoperability
Natively compatible with MCP, C2PA, and W3C PROV to seamlessly integrate into agent frameworks
Core Concepts
Evidence
A set of cryptographically verifiable claims, the atomic unit of the protocol
Artifact
Code, files, execution decisions, and data generated by AI agents
Event
Granular execution records, tool invocations, and environment context
Identity
Agent machine identity credential powered by AEP PKI
Evidence Graph
A causal graph of events and artifacts, fully capturing the agent's action chain
Blockchain Anchor
Evidence fingerprints anchored to global distributed ledgers, providing public existence and immutable finality
Evidence Levels
From L0 to L7, evidence strength increases progressively (cumulative)
PlainHash: any valid package payload (unsigned hash)
Signed: at least one valid asymmetric signature
Identity-verified: identity verified via CA/IdP certificate chain
Timestamped: high-precision trusted timestamp proof (RFC 3161)
Anchored: distributed global ledger existence anchoring
Notarized: endorsement by a neutral audit or attestation authority
Continuously-witnessed: continuous lifecycle execution records with multi-party consensus witnessing
Hardware-rooted: TEE / secure enclave hardware root of trust
Current Engineering Status: L0–L4 (Hash, Signature, Identity/Witness, Timestamp TSA, Blockchain Anchor Chain) are fully implemented and production-ready; L5–L7 are reserved for advanced protocol evolution.
Relationship with Other Standards
Evidence MCP
MCP is the de facto standard for agent-tool interaction; AEP treats Evidence MCP as a native core integration gateway to empower agents with native evidence capabilities.
C2PA
C2PA focuses on media content provenance (Content Credentials), while AEP concentrates on verifiable evidence chains across the full lifecycle of AI agent executions. The two are complementary.
W3C PROV
PROV provides a general provenance ontology specification. AEP's evidence graph closely aligns with its data model while being cryptographically optimized for AI behaviors.