What is Policy Rollback?

Policy changes can go wrong. A rule intended to block access to a sensitive database table might have a regex that's too broad, blocking all database queries. A new compliance rule might conflict with an existing operational rule, creating unexpected denials. A well-intentioned tightening of permissions might break a critical agent workflow that no one realised depended on the access being removed.

When a policy change causes problems, speed of recovery matters. If agents are being blocked from legitimate operations, every minute of downtime affects productivity or revenue. If a policy change accidentally permitted access that should be denied, every minute of exposure is a compliance risk. Rollback must be fast, reliable, and well-understood by the team.

Keeping a copy of each policy document in version control makes rollback straightforward. A git revert creates a new commit that restores the known-good JSON, preserving the full history (including the mistake and the fix). Pasting that document back through the Raw JSON view re-applies it, and the same save-time validation runs as for any other change. This is materially better than reconstructing a policy from memory, which has no record of what the known-good version actually was.

Because a PolicyLayer policy is a single JSON document attached to a grant, rollback is an edit operation — paste the known-good document into the Raw JSON view (or re-apply the previous settings in the visual policy builder) and save. There is no internal state to reset and no cache to invalidate: the document is validated on save and takes effect at the gateway on subsequent calls. Teams that keep exported policy JSON in version control can restore any prior version on demand. Recovery time is limited only by how quickly an operator can act.