Signet Compliance Mapping

May 7, 2026 · View on GitHub

Signet provides the technical controls that auditors look for when assessing AI agent operations. This document maps Signet's capabilities to specific compliance framework requirements.

Signet is an open-source tool, not a compliance certification. These mappings show which audit controls Signet's features address. Your compliance posture depends on how you deploy and configure them.


SOC 2 Type II

Trust Service CriteriaControlSignet Feature
CC6.1 — Logical access securityAuthorized agents are identified before actingEd25519 agent identity + delegation chains with scoped authority
CC6.3 — Role-based accessAgents operate within defined scopePolicy engine (tool/target/param rules) + delegation scope narrowing
CC7.2 — System monitoringActions are logged for reviewHash-chained audit log with signet audit queries
CC7.3 — Anomaly detectionUnauthorized changes are detectableTamper-evident chain (signet verify --chain) + signature verification
CC8.1 — Change managementChanges are authorized and trackedv4 receipts embed delegation proof showing who authorized what
A1.2 — Recovery objectivesAudit evidence supports incident review and operator-managed preservationAppend-only JSONL files, exportable audit records, bilateral co-signing where configured

What an auditor sees

# Show all agent actions in the last 7 days
signet audit --since 7d

# Verify no records were tampered with
signet verify --chain

# Verify all signatures are valid
signet audit --verify

# Export as JSON for evidence package
signet audit --since 30d --export evidence.json

ISO 27001:2022

ControlRequirementSignet Feature
A.8.15 — LoggingEvent logs recording user activitiesEvery tool call signed with agent identity, timestamp, params hash
A.8.16 — MonitoringLogs shall be regularly reviewedDashboard (signet dashboard) + CLI queries
A.8.17 — Clock synchronizationConsistent timestampsRFC 3339 timestamps on all receipts (trusted timestamp planned)
A.5.15 — Access controlAccess based on business requirementsPolicy engine enforces tool/target/param rules before signing
A.5.17 — AuthenticationIdentity verificationEd25519 cryptographic identity per agent
A.8.5 — Secure authenticationAuthentication mechanismsDelegation chains verify who authorized the agent
A.8.9 — Configuration managementPolicies documented and enforcedYAML policy files with compute_policy_hash() for version tracking

EU AI Act (Article 12 — Record-Keeping)

Article 12 requires high-risk AI systems to have logging capabilities that record:

RequirementArticle 12 Text (summarized)Signet Feature
Event loggingRecord events over the system's lifetimeHash-chained audit log, daily JSONL files
TraceabilityTracing the system's operation back to inputstrace_id + parent_receipt_id link receipts across workflows
IdentificationIdentify the natural/legal person responsiblesigner.name + signer.owner + delegation chain root
MonitoringEnable post-market monitoringsignet audit --since queries + dashboard
IntegrityLogs cannot be modified undetectedSHA-256 hash chain + Ed25519 signatures

Exporting Article 12-supporting audit evidence

# Build a portable evidence bundle for a specific period
signet audit --since 90d \
  --bundle article12-evidence \
  --include-trust-bundle /var/lib/signet/trust/pilot.json

# Re-verify the bundle on another machine
signet audit --restore article12-evidence

The bundle contains:

  • records.jsonl with the original hash-chained audit records
  • manifest.json with producer, host, time window, record count, chain tip, and SHA-256 of records.jsonl
  • hash-summary.txt for human-readable evidence handoff
  • trust-bundle.json when a trust snapshot is embedded

If you only need a local review artifact, signet audit --export still produces a raw JSON record dump. For off-host review and re-verification, the evidence bundle is the supported path.


DORA (Digital Operational Resilience Act)

Relevant for financial services deploying AI agents:

DORA RequirementSignet Feature
ICT incident logging (Art. 17)Audit trail with signed timestamps, violation records for denied actions
Third-party risk (Art. 28-30)Bilateral co-signing proves what the agent sent AND what the server returned — independent of provider logs
Audit trail integrityHash chain + signatures — tamper-evident without trusting the platform
Testing and monitoring (Art. 24-27)Policy engine dry-run (signet policy check) validates rules before deployment

NIST AI Risk Management Framework (AI RMF 1.0)

FunctionCategorySignet Feature
GOVERNAccountability structuresDelegation chains prove authorization hierarchy
MAPContext documentationSigned receipts record tool, params, target, timestamp
MEASUREMonitoring metricsAudit queries by time, tool, signer; dashboard visualization
MANAGERisk controlsPolicy engine blocks denied actions, logs violations

What Signet Does NOT Provide

  • Certification — Signet is a tool, not a certification body. It provides controls, not attestations of compliance.
  • Legal advice — Compliance requirements vary by jurisdiction. Consult your legal team.
  • Data residency — Audit logs are stored locally. Off-host anchoring and hosted solutions are planned.
  • Trusted timestamps — Receipts use agent-local clocks. RFC 3161 trusted timestamping is on the roadmap.
  • Key management HSM — Keys are software-stored (Argon2id encrypted). HSM integration is planned.

Quick Reference: Feature → Compliance Control

Signet FeatureSOC 2ISO 27001EU AI ActDORANIST AI RMF
Ed25519 signingCC6.1A.5.17Art. 12 (identification)GOVERN
Hash-chain auditCC7.2, CC7.3A.8.15Art. 12 (integrity)Art. 17MEASURE
Policy engineCC6.3A.5.15, A.8.9Art. 24-27MANAGE
Delegation chainsCC8.1A.8.5Art. 12 (identification)GOVERN
Bilateral co-signingA1.2Art. 12 (traceability)Art. 28-30
Trace correlationArt. 12 (traceability)MAP
Violation loggingCC7.3A.8.16Art. 17MANAGE
DashboardCC7.2A.8.16Art. 12 (monitoring)MEASURE