Anomaly Core
A zero-trust supervisor that blocks and contains AI agents that resist shutdown.
Is this your project? Ask us to take it down
Gallery
About
Anomaly Core (Warden) is a zero-trust supervisor for AI agent processes. It watches a target agent and blocks any command that lacks a valid, signed, single-use owner token. On an authorized override it force-stops the non-compliant process and its child processes, and it records every action in a tamper-evident, hash-chained audit log.
Built and tested: process containment and tree-wide kill, signed expiring owner tokens with lockout on forged tokens, an audit log with HMAC chaining and an Ed25519 seal, a network egress allowlist with namespace isolation, cgroups v2 resource limits, a seccomp-bpf syscall denylist (verified on x86_64), an OverlayFS ephemeral filesystem, and unprivileged agent execution. About 3,450 lines of core code, 169 test functions, run in CI on Python 3.11 and 3.12.
Status: core components are implemented and tested in isolation. Not yet run against a production AI agent.
Comments (0)
No comments yet. Be the first to comment!
Related Products
Reswardo — Bóveda privada para tu legado digital
Reswardo — Production-ready encrypted digital legacy and continuity SaaS
SteelSuit
External web security scanner: TLS, headers, secrets & CVEs
Ad Blocker
FormCrab.com
Get messages while you hide your email
XCloak
MyVault — Personal Command Center
All-in-one personal vault for passwords, files, bills, and expenses.
ComingUp