Identity at the edge
Sign-in is checked by Cloudflare Access (Zero Trust) before a request reaches FIRAX AI. Only people your administrators admit get in, with your company identity.
FIRAX AI puts identity, least privilege, human approval and an audit trail around every AI agent — and maps those controls to the frameworks your governance program already uses.
They apply to chat, apps, scheduled tasks and AI Studio workflows alike — there is no way around them.
Sign-in is checked by Cloudflare Access (Zero Trust) before a request reaches FIRAX AI. Only people your administrators admit get in, with your company identity.
Each person connects each system with their own account. A connection reaches only the resources chosen — one mailbox folder, one site, named tools — and nothing else.
Anything that changes data or reaches outside the company — sending, editing, creating — waits for a person to approve, edit or reject it.
Every call an agent or app makes through a connection is recorded with who asked, which agent, which system and what was done, together with the approvals.
Every app built in FIRAX AI runs in its own sandbox, private to its owner until it is deliberately shared. Credentials stay with the Gatekeeper; agents and apps never hold them.
FIRAX AI runs on Cloudflare's network in FIRAX's own Cloudflare account. Models are reached through Cloudflare AI Gateway, and administrators choose which models may be used.
The framework's four functions, what FIRAX AI provides for each, and what stays with your organization.
| Function | What it asks | What FIRAX AI provides | What stays with you |
|---|---|---|---|
| Govern | Policies, roles and accountability for AI use. | Administrators set the company instructions, the approved connections and models, and who may sign in. Every agent works under those rules. | Your AI policy, roles and acceptable-use rules. |
| Map | Know where AI is used, on what data, and in what context. | One inventory of agents, apps, workflows and connections, each tied to a person and to the systems it may reach. | Your list of use cases and their business context. |
| Measure | Track how AI systems behave and what they cost. | The audit trail of every call and approval; per-model usage, spend and limits in Cloudflare AI Gateway; test runs before a workflow is published. | Your review of outcomes and quality against your own measures. |
| Manage | Act on risks: limit, respond, and keep people in control. | Human approval for anything with side effects; scopes narrowed or connections removed at any time; published workflow versions that do not change. | Your risk decisions, incident response and periodic reviews. |
ISO/IEC 42001 asks an organization to run AI under a managed system of policy, risk, operation and review. These are the areas FIRAX AI's controls are designed to support.
Company instructions and approved connections and models put your AI policy into the system each agent works in.
Least-privilege connections and approval on every side effect give your risk assessment concrete controls to point at.
Draft, test, publish: AI Studio workflows are tested before use, and a published version never changes — a change is a new version.
The audit trail and model usage records are the evidence your internal reviews and audits read.
People stay accountable: every change is approved by a named person, and every answer can be traced to the call that produced it.
Found a vulnerability? Write to support@firax.ai with the details. Evaluating FIRAX AI? We will walk your security team through the controls on a discovery call.