Sentinel

Autonomous coding agents like IBM Bob represent a transformative leap in developer velocity, autonomously writing source code, executing raw shell commands, altering infrastructure, and committing directly to repositories; however, enterprise adoption faces a critical barrier: the absence of verifiable trust and active safeguards, since today, agent telemetry resides in passive, transient logs that can be silently altered or deleted by rogue actors or prompt injection attacks, and if an agent hallucinates a destructive command or modifies production credentials, existing monitoring tools only record the disaster after the damage is done, meaning enterprises cannot risk deploying root-level autonomous AI without provable accountability. To address this, we built Sentinel, a zero-trust AI Governance and Cryptographic Audit Engine designed specifically to secure IBM Bob workflows, tailored for enterprise SecOps teams, compliance officers, and engineering leads, transforming unmonitored agent autonomy into a transparent, compliant, and mathematically provable operational pipeline. Users monitor live Bob execution sessions on Sentinel's high-fidelity incident dashboard, tracking active runs, event throughput, and real-time risk scores, and when an agent attempts a sensitive or high-risk operation-such as deleting core configuration files or executing privileged shell scripts-Sentinel's policy engine halts execution and routes the event to a Human-in-the-Loop (HITL) approval queue, where a human reviewer inspects the blast radius and cryptographically signs off before Bob can proceed; through the interactive Timeline and Tamper Lab, operators inspect forensic payloads and execute one-click mathematical audits across the entire historical chain. Unlike conventional observability platforms that merely report historical failures, Sentinel introduces two novel architectural innovations: first, Active Pre-Execution Interception, where Sentinel does not passively record incidents after they occur but functions as an inline circuit breaker, intercepting tool invocations and physically enforcing human control gates before destructive code executes on the host; and second, PostgreSQL SHA-256 Hash Chaining, where instead of relying on vulnerable standard database records or slow, complex blockchains, Sentinel chains every single event into an unbroken, tamper-evident sequence [H(n) = SHA256(H(n-1) + Payload)] directly within PostgreSQL, so that if anyone-even an internal database administrator-silently alters an event byte, Sentinel's verification engine instantly flags the broken link and pinpoints the exact corrupted event ID. By uniting active human oversight, automated policy enforcement, and mathematical proof of immutability, Sentinel satisfies SOC 2, ISO 42001, and EU AI Act mandates, giving organizations the confidence to deploy IBM Bob at enterprise scale.