Trust and Security

Built to pass your security review.

We know AI suppliers to regulated organisations face the hardest questions in procurement. Here are our answers, in plain language.

Our compliance language commitment. We describe our alignment with regulatory frameworks precisely. We do not claim endorsement or approval by any regulator, and we will never describe a capability as certified unless the certificate is held. Where guidance is in draft, we say so.
Security by design

Designed controls, layer by layer.

Network

Deployment inside your network segment or private cloud. Default deny egress at firewall level. No telemetry and no call home. Egress rules verified at installation and monitored continuously.

Identity and access

Single sign on through your identity provider (Entra ID, SAML or OIDC). Role based access. Least privilege service accounts. Electronic signature binding designed in alignment with 21 CFR Part 11 and EU GMP Annex 11.

Data

AES-256 encryption at rest, TLS 1.3 in transit, customer held keys where required. The corpus never leaves your environment.

AI specific threats

Retrieved content treated as untrusted. Separation of instructions and data. Every output validated by the deterministic engine before release. Jailbreak attempts logged as incidents. Model files hash verified at deployment.

Supply chain

Software bill of materials, dependency scanning, signed containers and reproducible builds. Vendor assessment pack for your security review.

Platform assurance

GAMP 5 lifecycle. Penetration testing before first production deployment and annually thereafter. Tamper evident audit trail.

Responsible AI principles

Six commitments, enforced in software where they can be.

01

Accountability stays human

AI assists. Named people decide and sign.

02

Evidence before reliance

Nothing is used before it is recorded.

03

Consequence sets the control

The higher the stakes, the deeper the evidence.

04

Sovereignty by default

Your data, models and records stay under your control.

05

Transparency about limits

We are clear about what our systems do not do, and enforce exclusions in software.

06

No training on your data

Your information is never used to train or fine tune models.

Certifications and assurance documentation

Our security certification roadmap and vendor assessment pack, including our software bill of materials approach, penetration testing policy and platform risk assessment summary, are available to organisations under NDA.