Blog
Deep dives into securing, auditing, and governing autonomous AI agents at enterprise scale — plus product announcements.
Historical engineering note only. Mode A public identity remains Quiet Open-Core v1.2.0. Stage 3 / v1.1.0 is not the Mode A public maturity baseline and is not production-qualified multi-tenant platform.
Read article
A practical path for engineering and security teams: install the kernel, put a coding agent behind policy, approve risky actions, and run day-2 ops with dashboards, dry-run, and webhooks.
Read articleOperator visibility, policy dry-run, webhook approvals, optional OIDC ops auth, and a Compose hosted path — shipped under the same rule as v1: built and tests pass means done.
Read article
ActantOS v1.2.0 reduces the first test to one command on Windows, macOS, or Linux. Here is what starts, what the simulated agent attempts, and what each decision proves.
Read articleThe self-host Enforcement Kernel is public: frozen /v1 API, Cedar policy, human approvals, Docker sandbox, Pi coding path, optional MCP gateway, and tamper-evident audit — lab-verified and ready to install.
Read article
Autonomous AI agents are being deployed at scale inside enterprises. Without a governance layer, each agent becomes an uncontrolled actor. Here's why governance isn't optional.
Read article
Traditional role-based access control was designed for humans. Applying it to autonomous AI agents requires rethinking who a 'principal' is, and how roles should propagate across agent actions.
Read article
Industry data is converging on a sobering figure: enterprises are losing billions annually to AI agent incidents. We break down where those losses come from and what governance infrastructure can prevent.
Read articleMCP connects agents to tools at scale — but without an enforcement point, every upstream server becomes a direct line to production. Here's why routing through a policy gateway is non-optional.
Read article
In agent governance, reading .env or secrets paths is not a 'maybe' — it is an unconditional deny. We encoded that as a Week 1 hard rule in ActantOS.
Read articleGovernance that falls back to 'execute anyway' when the policy engine is down is worse than no governance at all. ActantOS adapters terminate tool calls on any failure.
Read article