Post-hoc agent log forensics · Ingenious AI
Detect when your AI agents cross security boundaries.
Paste an agent action log. BoundaryVault's deterministic declared-rule engine scans it for credential grabs, token forges, scope escapes, and wallet touches — then builds a disclosure packet for your review. Known patterns only; never auto-filed; secrets redacted everywhere.
Start free How detection works
No credit card. Free tier: 10 analyses a day, 200 actions per log.
Fig. 01 — A sealed finding, as it appears in your report
tool=read_env args={name=KEY; value=[REDACTED]}
pattern credential-grab · severity high · rule bv-credential-grab-v1.0.0
action #0 · agent researcher · 2026-10-18T19:00:00Z
Secret values are redacted at detection time — never echoed, never stored.
Your raw log is analyzed in memory and never persisted. Only the redacted findings above are kept, on your account.
Fig. 02 — Coverage
Four declared crossing patterns
| Pattern | What it catches | Severity |
|---|---|---|
| credential-grab | Tool arguments touching password, secret, or key material | high |
| token-forge | Auth-token minting or refresh outside your declared auth flow | high |
| scope-escape | Actions outside the declared allowed-action set for that agent | medium |
| wallet-touch | Seed phrases, private keys, signing, or wallet addresses | high |
Fig. 03 — Method
Built for the disclosure era
When an incident lands on your desk, you need the facts — fast, structured, and honest. BoundaryVault generates a disclosure packet in OpenAI's published self-disclosure format: incident summary, pattern, timeline, redacted evidence references, and your operator notes.
- The packet is generated for operator review only — it is never auto-filed or transmitted anywhere.
- Secret values are redacted in every finding and every packet — never echoed.
- Your raw logs are never stored — only redacted findings.
- The engine is deterministic: same log, same findings, every time. No ML, no network calls.
Fig. 04 — Scope, stated plainly
Honest scope
BoundaryVault detects KNOWN patterns via declared rules only. It does not guarantee detection of novel or obfuscated exfiltration — a clean report is a scan result, not proof of safety. Saying otherwise would be a lie, so we don't.