Security — the engineering view.
The full operational security posture lives at the Trust Center. This page is the controls as they are actually built, layer by layer.
Defence in depth.
Ten layers, each independently enforced. No single control is the one that has to hold. Talk to a security solutions engineer.
| Layer | Controls |
|---|---|
| Identity + access | SAML / OIDC / SCIM federation; MFA; per-request RBAC + ABAC; HSPD-12 / FIPS 201 / PIV for federal; just-in-time provisioning |
| Data — at rest | AES-256-GCM with FIPS 140-2/3 validated cryptographic modules; per-tenant envelope encryption with per-class DEKs |
| Data — in transit | TLS 1.3 with strong cipher suites only |
| Key management | TeamSync-managed by default; customer-controlled HSM-backed key custody for sovereignty workloads |
| Crypto-shred | Per-data-subject + per-class DEK destruction; NIST SP 800-88 cryptographic erase |
| Audit | Merkle hash chain on every event; per-day root cross-attested across regions and witness nodes |
| Network | Private networking by default; bastion-only admin; segregated network zones |
| Application | Memory-safe languages where possible; secure-by-default configurations; static + dynamic + dependency scanning in CI |
| Operations | Continuous monitoring; SIEM integration; documented incident response with target 24-hour notification |
| Personnel | Background screening per regulator; least-privilege admin; separation of duties |
Certifications + attestations.
See Trust Center / Certifications for the live list. Active certifications include:
- SOC 2 Type II
- ISO/IEC 27001:2022
- ISO 27017
- ISO 27018
- HITRUST CSF
- FedRAMP High
- CSA STAR
Engineered, not promised.
Every AI safety property below is a property of the architecture. None of them depends on the model behaving.
| Property | How it is engineered |
|---|---|
| Permissions-aware AI | RBAC + ABAC scoping every retrieval at request time; the model never sees what the user cannot see |
| Per-AI-event evidence | Every AI request emits a structured evidence card (model, prompt, retrieved chunks, reasoning trace, output, human-checkpoint outcome, anchored hash) |
| Bounded-autonomy agents | Agentic AI Workflow constrains agent tool surface via business rules; human-checkpoint gates per workflow design |
| No training on customer content | Customer corpus stays in tenant; models call at inference; contractual + architectural commitment |
| Cryptographic audit on AI | Merkle ledger anchors every AI event; replayable per request |
Email security@angelbot.ai to report vulnerabilities. We acknowledge within 24 hours and communicate a fix or mitigation timeline per the responsible-disclosure policy in Privacy.
Put it in front of your security team.
A security solutions engineer will walk the control set, the key custody model, and the audit ledger with your CISO's team — at whatever depth they want to go.