R&D-001 extension
Agent Security & Resilience
Research into the controls enterprises need beyond basic AI governance: non-human identity, tool/MCP trust, runtime enforcement and recoverable agent state.
Tool & MCP trust
Allow-lists, declared artifact pins and resolved-checkout identity checks to reduce tool supply-chain risk.
Agent state resilience
Integrity-checked state snapshots and recovery planning after corruption or drift.
Incident evidence
Structured internal incident ledger and deterministic triage for unauthorised actions, safeguard failures, recurrence and external impact.
Current research prototype v0.3: 9 automated tests passing. Public summary only.