How Penaxtra protects AI security data, credentials, audit logs, webhook secrets, runtime gateway policy, and tenant isolation. Each layer specified with the same control-ID precision we ship to your auditors.
Modern cipher suites only; HSTS preload locks the certificate path once the production domain is fixed.
Five distinct database roles segregated by workload, each with the minimum grants its function requires.
Row-level access control enforced at the database layer; isolation context is reset on every connection checkout.
Tamper-evident audit channel plus a writer role with INSERT-only grants. Audit retention is tenant-configurable up to ten years.
Procurement reviewers can verify each layer against a control identifier in our compliance matrix.
TLS 1.3 only at the public edge. Web application firewall, distributed denial-of-service mitigation, and origin certificate pinning enforced. Content-Security-Policy is strict (no inline scripts, no inline styles, nonce-based exceptions only).
Time-based one-time passwords (RFC 6238) with ten one-time recovery codes, plus optional passwordless magic-link sign-in (15-minute single-use, bound to IP + user-agent fingerprint). Organization-enforced MFA is planned for Enterprise identity controls. User-level TOTP is shipped today.
Database-layer row-level access control enforced on every tenant-scoped table. Five distinct database roles segregated by workload (web, backend, migrations, read-only analytics, backup) with least-privilege grants. A connection pool resets the tenant isolation context on every checkout.
Authenticated public-key encryption (X25519 sealed-box AEAD) for every credential at rest, including LLM provider tokens, OAuth tokens for chat integrations, ticketing API credentials, SIEM tokens, and customer webhook signing secrets. Reset tokens are SHA-256 hashed; the raw value exists only in the user inbox.
Append-only audit log. The writer role has INSERT-only grants; UPDATE, DELETE, and TRUNCATE are revoked. A tamper-evident database audit channel mirrors every event to a separate sink so an actor with application-level access cannot rewrite history. Retention is tenant-configurable from one day to ten years.
Nightly database dumps compressed with zstd and encrypted with age (X25519 recipient), shipped to a private object-storage bucket. Seven-day local retention; off-site retention at least 90 days. Report PDFs receive an additional age-encrypted off-site copy.
Scan-target output is never re-introduced into the judge prompt as instruction. Every adversarial probe is templated; user-controlled content is parameterized, never string-concatenated into a directive.
Outbound webhooks must resolve to a publicly routable IP. Loopback, link-local, RFC 1918, and multicast targets are rejected at fire time. Hostname-IP resolution is re-checked at the moment of delivery to defeat DNS rebinding.
Judge rationales pass through a redaction stage before persistence. Credentials, payment numbers, and personal identifiers are replaced with class tokens before they hit the database.
Enterprise identity, access-control, and admin-side observability features below are tracked on the public roadmap. Items marked planned are not available in the current platform release; we never market a planned feature as shipped.
User-opt-in TOTP plus backup codes. The platform never enforces MFA at the org level by design; the option is exposed to every user in account settings, with backup codes minted at enrollment.
Database-layer row-level access control on every tenant-scoped table; five segregated database roles by workload (web, backend, migrations, read-only analytics, backup).
X25519 sealed-box AEAD for tenant secrets. Cleartext never lands on disk.
HMAC-signed delivery with timestamp + replay protection. Customer-supplied secret per webhook target.
Mirrored to a tamper-evident audit channel at the database layer. Configurable per-tenant retention from one day to ten years.
Ed25519-signed policy bundles to runtime gateway agents; agent verifies before applying.
Enterprise tier addition. Roadmap entry: Q4 2026. Pre-orderable under enterprise contract.
System for Cross-domain Identity Management for automated user lifecycle. Pairs with SSO/SAML roadmap.
Per-tenant restriction of console + API access to a CIDR allowlist. Roadmap entry.
Scheduled CSV / SIEM HEC export of platform admin actions. SIEM forwarder for tenant events is shipped today; admin scope is roadmap.
Enforced rotation cadence for API tokens with grace-window warnings. Roadmap entry.
WebAuthn / FIDO2 second factor. Roadmap entry; backup codes today.
Email [email protected] or follow the machine-readable contact at security.txt (RFC 9116). Encrypted reports preferred (PGP key linked from security.txt). 90-day coordinated disclosure window.
CVSS 4.0 and the OWASP Risk Rating method. Critical findings trigger a 24-hour patch SLA. High issues are patched within 7 days, medium within 30 days, low at the next release.
A monetary bounty programme is in scoping. Today we acknowledge contributors on a public thank-you wall at security.txt and never pursue legal action against good-faith research conducted under the disclosure policy.
Every customer-provided credential (LLM tokens, OAuth tokens, integration API keys, webhook signing secrets) is encrypted at rest with X25519 sealed-box authenticated encryption. The encryption key lives outside the application repository in a restricted-access configuration store. The application decrypts on demand and never logs decrypted material.
Same intake: see security.txt for the machine-readable contact. The gateway agent is signed; if you suspect signature bypass, please include reproduction artifacts. We will issue a CVE and a coordinated update to all enrolled customers.
Architecture diagrams, trust boundaries, signed-rule format, and a sample audit-evidence export, in one PDF.