Security Brief

Summary

Ainvento designs, builds and operates AI systems for government and regulated enterprise. Each client environment is isolated, encrypted with client-held keys, and deployed in the jurisdiction the client’s data classification requires — a Canadian cloud region, the client’s own data centre, or an air-gapped enclave.

Data handling

• Client data is processed only on documented client instructions and for the purpose of the engagement.
• Client data, and anything derived from it, is never used to train models for any other client.
• Data is classified at intake and handled according to the client’s security categorisation.
• At the end of an engagement, data, weights and indexes are returned in open formats and deletion is certified in writing.

Hosting and residency

Managed deployments run in Canadian cloud regions. On-premise and air-gapped deployments run on client infrastructure. Inference, logging and backups stay in the same jurisdiction as the primary data; no component calls an external model API unless the client approves it in writing.

Access control

• Single sign-on with enforced multi-factor authentication for all staff.
• Role-based access on least privilege, reviewed quarterly.
• Production access is just-in-time, approved per request, time-limited and logged.
• Personnel with access hold the security clearance the environment requires.

Encryption

TLS 1.2 or higher for all data in transit. AES-256 for data at rest. Keys are held in the client’s key management service or hardware security module, and Ainvento can operate the platform without access to them.

Logging and monitoring

Every model output is logged with its input, the sources it retrieved and the model version that produced it. Administrative actions are logged separately. Logs are retained to the client’s schedule, exportable on request, and monitored for anomalous access.

Model governance

• Each model release records its base model, training data snapshot and evaluation results.
• Releases pass a client-owned evaluation set and a red-team review before deployment.
• Clients are notified before any model change and may hold a pinned version.
• Algorithmic Impact Assessment inputs are provided for systems that support administrative decisions.

Incident response

Ainvento maintains a documented incident response plan, tested at least annually. Clients are notified of any confirmed incident affecting their data within 24 hours, with a written report following resolution.

Business continuity

Managed environments are backed up daily to a second location within the same jurisdiction, and recovery is tested at least twice a year. Recovery objectives are agreed per engagement and written into the service agreement.

Assurance

Ainvento’s certifications, audit reports and penetration test summaries are listed on the Trust & Compliance page and are available to evaluation teams under NDA.