You built the agent. You provisioned the credentials. Nobody documented what it was actually authorised to do.

When the agent was deployed, the service account was provisioned with the access it needed to work. Nobody compared that access against the business policy that was supposed to constrain it. The gap has existed since day one.

The declared authority: what the CFO or product team said the agent could do, lives in a Confluence page, a Slack thread, or nobody's memory. It has never been reconciled against what the service account actually permits.

Every time a new system is connected to the agent, the technical authority expands. The declared authority does not automatically update to match. The compliance and security teams do not always know a new connection was made.

When the security team asks for an audit trail of a specific agent action, you can pull the logs. But the logs show what happened inside your systems. They cannot independently prove the organisational authority behind what happened, and that is the question the enterprise buyer's security team is now asking.

Sub-agent delegation is invisible to your observability stack. Your monitoring tells you what each agent did. It does not tell you what financial authority propagated through the delegation chain.

A structured connected system inventory: every system your agent can access, the credential it uses, and the technical permission it holds, produced from your own systems rather than from memory.

A reconciliation of your IAM configuration against the declared business policy, showing specifically where technical access exceeds what was organisationally approved and what needs to change.

An Action Replay for specific past transactions: giving your team an independently reconstructable record of what happened, useful for incident investigation and post-mortems.

A remediation plan with specific technical instructions, which service accounts to restrict, which permissions to scope down, which approval gates to implement at the system level rather than the prompt level.

Authority Watch: automated monitoring that alerts your team when permissions change, new systems connect, or technical authority expands beyond what was documented in the assessment.