Federated Threat Intel
Opt-in cross-tenant attack signature sharing. When an attack is detected on one tenant, a privacy-preserving fingerprint is contributed to a shared pool — and every other opted-in tenant gets the protection before they ever see the attack themselves. Off by default; enable in Settings → Federation.
Features
Privacy-preserving by construction
Original prompts are never stored. Contributions are HMAC-SHA256 fingerprints over normalized attack patterns; the raw text and the tenant identity are dropped at the edge.
PII gate before hash
A pre-hash PII detector blocks any candidate signature that still contains emails, SSNs, phone numbers, API keys, or other identifiers — so the pool never ingests sensitive payloads even by accident.
k-anonymity promotion threshold
A signature is only promoted into the active distribution pool after at least 3 distinct tenants have independently contributed it. Single-tenant or low-quorum patterns never leave the staging table.
Consent-versioned opt-in
Operators enable federation per-tenant in Settings → Federation; consent is stamped with a version and audit-logged. Disabling immediately stops future contributions and cuts the tenant off from receiving matches against the pool. Previously-contributed signatures stay in the pool by design — they're already anonymous and other tenants still benefit from them.
Distribution to scan path
On every prompt-injection check, the scanner runs an additional step against the promoted federated pool — independent of, and in addition to, the static OWASP catalog and managed ruleset. A match contributes to the same risk score as the built-in detectors.
Operator visibility
Admins see the active pool, contributor count per signature, and recent promotions at /mgt/federation. Per-tenant operators see their own contribution + consent state in Settings → Federation.