Aigentical
Security & Trust

Built to be handed to your security team.

Aigentical sits in the path of your AI agents' actions, so we hold it to the standard that placement demands: single-tenant, your keys, a record you (not we) control, and nothing that leaves your boundary unless you send it there.

Last updated: October 2026 · security@aigentical.ai

Where it runs

  • Your infrastructure. Aigentical deploys into your cloud, on-prem, or air-gapped environment — a single-tenant control plane, not a shared multi-tenant SaaS.
  • Your boundary. Governance decisions are made inside your network: nothing is routed to us to make a decision, and the policy engine runs where you deploy it. What leaves your network is only what you configure — for example, an allowed request the LLM proxy forwards to the model provider you choose.
  • Any surface. Insert as the LLM proxy, the Actions API & SDK, MCP intercept, or the RAG firewall in front of retrieval — without re-architecting your agents.

What we store

Governance decisions and their metadata — the acting identity, the action, the verdict (allow / hold / deny), the reason, and a timestamp — written to a tamper-evident audit chain.

the record · what gets written
acting identitywho acted: the resolved agent identity
the actionwhat it tried to do
the verdictAllowHoldDeny
the reasonwhy it was decided
timestampwhen
◆ written to a tamper-evident audit chain
  • Not your model content by default. Prompts and completions are scanned in-path for policy, but the audit record captures the decision and its rationale, not a copy of your data — and you control what is retained and for how long.
  • We do not train on your data. Your traffic, policies, and records are yours; they are never used to train shared models.

Your keys, your custody

BYOK

The signing key is yours

The audit chain is HMAC-linked, and the signing key is a secret you hold in your own deployment — we never hold it. Because the key is yours, the record can't be silently rewritten, including by us, and any tampering shows the next time the record is verified: in the console, or with the standalone verifier you can run yourself.

Secret handling

Never bundled, never in source

Integrity keys and secrets are never bundled with the application or checked into source.

Identity

Identity — we consume, we don't replace

  • Aigentical sits on top of the IAM you already run: SPIFFE / SPIRE, mTLS, OIDC & SAML SSO (Okta, Entra, Keycloak), and SCIM provisioning.
  • Under enforce, a call without a cryptographic agent credential is refused, and dev-scoped tokens can't act in production. In monitor mode such calls are let through so you can find every agent, and each is recorded as weakly identified — never as proven.

Compliance posture

Our controls are mapped to the frameworks your program already answers to:

SOC 2CC6 / CC7
EU AI ActArt. 12 & 14
ISO42001
NISTAI RMF
HIPAA§164.312
PCI-DSSReq. 10
GDPRArt. 30 & 32
SOC 2: not yet certified.
Our controls are mapped to CC6 / CC7 today, and formal SOC 2 attestation is on our roadmap as we mature alongside our first enterprise customers.

Framework mappings are Aigentical's own and are provided to accelerate your program; they do not imply certification or endorsement by the framework owners.

Reporting a vulnerability

Found something? Email security@aigentical.ai. We'll acknowledge quickly and work the fix with you; we don't pursue good-faith researchers.

security@aigentical.ai