Policy as code
Policy as code means writing rules, such as access rules, as text a machine can check, under the same review and change control as software. The rule a reviewer reads is the rule the system applies.
In a run
Ares applies the access policy on each call, not the agent. When a call falls outside it, the refusal and its reason go on the record, so each rule can be compared with what it did in a real run.
Our word for it
An access policy you can read.
Why it matters for agents
Content an agent reads can change what it tries to do. A rule written into the agent’s own instructions can be overridden by that content. A rule applied outside the agent, on the path to the tool, holds whatever the agent now wants. That is the case for keeping agent rules as policy, apart from the agent.
The practice comes from software teams. It fits agents for the same reason: a machine checks the rule every time, so no one has to remember it.
Reading the rules, and testing them
A policy helps only if the people who answer for it can read it. On Ares, the policy is set with your team, and each refusal on the record names what was outside permission. Run a workflow in the sandbox or in the evaluation, and the refused step shows the rule at work.
Where it stops
- Policy as code is a way to write and review rules. It does not make a rule right. You approve each one.
- A rule has no effect on an agent that runs outside Ares.
Related terms
- Access policyAn access policy is the set of rules for who and what each AI agent may reach, and what it may do there.
- AI governanceAI governance is the set of rules, roles and checks an organisation uses to decide how AI may be used.
Tell us about one workflow.
We will say whether an agent should do it, and what the first step costs.
A senior member of the team replies within two business days.