Christophe Hamchin

AI runtime governance architecture brainchild

By day: programme delivery lead (conventional enterprise software) in regulated environment (telco) . By night: applied security researcher building runtime AI governance for safe, affordable agent deployment at scale, aligned with Singapore AI Safety Hub's mission for trustworthy AI.

The challenge: static policies and paper GRC cannot stop runtime injection or enforce governance against autonomous agents that reason in real time. The solution: dynamic execution-level enforcement of agent containment via a [1+3] Defensive AI-Security Runtime Governance Architecture: Sidecar observability plane (load-bearing layer): out-of-band OpenTelemetry anchored in hardware-isolated substrates (Hardware Security Module, Data Processing Unit, Baseboard Management Controller) to ensures tamper-proof evidence, even under kernel compromise.

Three decoupled enforcement tiers: - Tier 1: supply chain integrity via cryptographic verification and immutable storage - Tier 2: reasoning layer proxy with semantic filtering and real-time validation - Tier 3: zero-trust action layer with cryptographic identity and ABAC at API boundary

Scales from resource-constrained organisations (Tier 3) to regulated enterprises (all tiers). Delivers cryptographic attestation, active boundary enforcement, and hardware-anchored isolation for safe agent autonomy.

180-page strong document describes theory and operationalisation. Experimental results by Q4 2026.