Trust

Certifications, security posture, and evidence

What AI-FW is certified for, how it handles your data, and where your reviewers can verify both independently.

Certifications

Both certifications are published in the IAF CertSearch database, the global register of accredited management system certifications. Verify each record directly rather than taking our word for it.

ISO/IEC 27001

Information security management

The international standard for an information security management system: how an organisation identifies, manages and continuously improves information security risk.

Verify in IAF CertSearch

ISO/IEC 42001

AI management system

The international standard for managing AI responsibly: governance, risk treatment, transparency and lifecycle controls for artificial intelligence systems.

Verify in IAF CertSearch

The IAF CertSearch record is the authoritative source for certificate scope, certifying body and validity dates.

Security posture

Four properties that decide most security reviews before features are even discussed.

Runs in your environment

The gateway is a container in your own network, in front of your existing identity stack. Prompts are inspected locally, with no third-party service in the data path.

Data residency

Fail-closed by design

If inspection cannot run, the request is refused rather than forwarded unscanned. Availability policies (retries, failover) never become a route around policy.

Guardrails guide

Metadata-only audit

Transactions record identity, model, decision, rule, timing and outcome with truncated previews. Raw prompt and response content is not persisted, so there is no content store to breach.

Audit and export

Evidence your reviewers can use

Every decision is attributable to an identity and exportable to your SIEM over encrypted syslog or a pull API, so audit evidence lives in the systems you already monitor.

Audit Logs reference

How data is handled

  • Prompt and response inspection runs inside the deployment, not via a hosted classification service.
  • Provider credentials are held server-side in the model registry and injected on forward, so endpoint agents do not carry model keys.
  • Agent identity uses certificates you control, so revocation is per agent rather than a fleet-wide secret rotation.
  • Transaction records are metadata and truncated previews only; retention is configured rather than implied.
  • Export goes to your own logging platform, which means the durable copy of the evidence is in your control.

Deployment specifics, including the database backend and network placement, are covered in the deployment recipe and the settings reference.

Reporting a vulnerability

Report suspected vulnerabilities to our contact channel (or support@aifw.io) with reproduction steps and the affected version. We will confirm receipt, keep you informed through triage, and credit you in the release notes if you would like. Please do not test against systems you do not own.

Need a security review pack?

Ask for the Technical Plan: a step-by-step proof of concept you run in your own environment, so your reviewers can validate the posture above against live traffic rather than a document.