AI-AGENT SECURITY

Guardian

guardian.cam

Runtime containment for AI agents, enforced from outside the agent process. An external watchdog that decides whether the agent may act at all — not a policy doc, not an SDK the agent can ignore.

• Policy engine — declarative boundaries per agent: files, processes, network endpoints. Violations denied before execution.
• Process sensors — CPU, RAM, threads, file and network access: OS-level truth, not agent self-report.
• Kill switch — real SIGSTOP/SIGCONT containment; the watchdog freezes the process even if the agent is compromised.
• Hash-chained audit ledger — tamper-evident sha256 chain over every decision; court-grade evidence.
• Sovereign + Gauntlet — on-prem/air-gapped edition, and a public 211-test adversarial battery: 211/211, zero holes.

Visit guardian.cam →
← All BTTsoftware products