Trust & Security

Security at AuthRAI

We handle authentication and authorization for your AI agents. Security isn't a feature — it's the product.

Ed25519 Signing

Every agent token is cryptographically signed with Ed25519. Verification requires no database lookup — the signature is the proof.

SHA-256 Chain Audit Log

Every event is hash-chained to the previous one. Any retroactive tampering with audit history is cryptographically detectable.

Short-Lived Credentials

Tokens expire in seconds to hours. No standing credentials means a compromised agent's blast radius is bounded by the TTL.

Technical security controls

Timing-safe authentication

Bcrypt comparison runs for unknown emails too — prevents user enumeration via response timing.

RBAC with 5 role tiers

viewer → auditor → developer → admin → owner. Each role enforced at the API layer, not just the UI.

Per-request CSP nonces

Content Security Policy headers with unique nonces per request — mitigates XSS even if an injection point exists.

Rate limiting on every auth endpoint

Login, signup, password reset, and token issuance are individually rate-limited by IP with Redis sliding windows.

HMAC-verified webhooks

Every outbound webhook includes a X-AmpGate-Signature header. Verify before trusting delivery.

HSTS + TLS 1.3

All endpoints served over TLS with HSTS enforced. No HTTP downgrade possible once a browser has visited.

Delegation depth enforcement

Agent sub-delegation chains have a hard maximum depth. A compromised agent can't escalate privileges by chaining delegations.

Session invalidation on password reset

Resetting a password immediately revokes all existing sessions — stolen session cookies become useless within seconds.

Found a vulnerability?

We take security reports seriously. Please disclose responsibly — don't publish details until we've had a chance to fix it.

Report a vulnerability →

We aim to respond within 24 hours. Critical findings may be eligible for recognition.