Skip to main content
AGA v2.8.0

Capability Matrix

19 capabilities where AGA provides cryptographic proof. For each row, we document what exists today, not what is planned.

UNKNOWN is used where public documentation does not confirm or deny the capability.

CapabilityAGAAnthropicMS AGTStrataMCP Proxy
Sealed policy artifact
Continuous external enforcement
PARTIAL
Signed receipts per decision
Tamper-evident chain (hash-linked, monotonic)
Offline verification (no network, no API)
Chain continuity verification
Timestamp monotonicity verification
Cryptographic policy binding to sealed artifact
Response hash attestation
Independent decision re-derivation
Phantom execution with timing-matched spoofing
Threshold signing (Shamir, k-of-n)
Post-quantum hybrid signatures (Ed25519 + ML-DSA-65)
Roughtime timestamp corroboration
Network isolation verification receipts
Portal self-attestation (binary SHA-256)
Air-gapped operation
PARTIAL
Third-party verifiable without trusting operator
Open-source independent verifier

Notes

Anthropic Managed Agents: First-party governance. Anthropic defines policies and enforces them on their own models. Logging is internal. No independent third-party verification is architecturally possible because the governed entity and the governance provider are the same organization.

Microsoft AGT (AutoGen Team): Agent framework with tool-use orchestration. Governance is application-level, not cryptographic. No sealed artifacts, no signed receipts, no offline verification.

Strata Identity: Identity governance platform. Focuses on identity federation and access control, not AI agent runtime governance. No tool-call-level receipts or evidence bundles.

Standard MCP Proxy: Generic MCP protocol proxy. Forwards tool calls without governance evaluation. No policy enforcement, no receipts, no verification.