Security and compliance for AI agents
Every agent gets its own identity, governed interactions are inspected and recorded, and the audit trail is append-only — signed and hash-chained under the default configuration, which an operator can turn off.
Mapped to the standards that govern AI
The runtime controls and the evidence behind them.
SOC 2
Access control, audit logging, and monitoring evidence auditors look for. →
GDPR
Data-subject rights and erasure for AI systems. →
EU AI Act
Article 12 log-keeping and Article 73 incident reporting, in the record Praesidia produces automatically. →
ISO/IEC 42001
An AI management system you can operate and evidence. →
NIST AI RMF
Govern, map, measure, and manage — applied to autonomous agents. →
OWASP LLM Top 10
The agent-level risks — and the controls that address each one. →
Security built into every layer
Not bolted on after the fact — the platform is the security.
- Identity for every agent
- Each agent authenticates with its own credential — attributable, individually revocable, no shared service accounts. Revoking one agent contains the blast radius to that agent.
- Append-only audit trail
- Praesidia records the decisions it makes in an append-only, hash-chained audit trail, signed under the default configuration; an operator can turn signing off. An exported trail can be checked independently, offline. External anchoring is optional; without it, removal of the most recent records cannot be detected.
- Tenant isolation
- Each organization's data is isolated from other organizations'.
- SSO and SCIM for the humans
- For the people who manage your agents: SAML and OIDC single sign-on, SCIM user lifecycle, MFA, and passkeys.
- Least-privilege access
- Role-based access control and scoped API keys: every human and integration gets the access it needs and no more.
- Content guardrails
- Bidirectional inspection blocks prompt injection and stops sensitive data from leaving through an agent's output.
- Signed webhooks
- Outbound events are signed, so your systems can verify they came from Praesidia before acting on them.
- Vaulted secrets with BYOK
- Provider keys and agent credentials are vaulted — shown once, never read back in plaintext. Regulated deployments hold their own signing key (BYOK) for audit records.
How the planes fit together: architecture. Control status, hosting and attestations: Trust Center.
Don't take our word for it — verify it yourself
Every exported compliance bundle can be checked by your team or your auditor with @praesidia/audit-verifier — an Apache-2.0, dependency-free CLI that runs fully offline and never calls our servers, available from your Praesidia contact.
It confirms every signature, the full hash chain, and every tamper check — plus the Sigstore Rekor anchor when external anchoring is enabled. It also states what it does not prove.
A European company, GDPR by default
Security starts with who processes your data and on what terms. Praesidia is operated by an EU-incorporated company under GDPR, with data-subject rights and erasure built into the platform.
Current processors are listed on the subprocessor page; background on the wider topic is in data residency for AI agents.
Found a vulnerability? Our vulnerability disclosure policy explains how to report it and how to keep testing safe; contact details are at /.well-known/security.txt. We also test ourselves.
Everything a security review asks for, on one page
Control status, hosting and data location, attestations, and the security contact.