# Autonomous Remediation — Agents, Approvals and Audit Trail | Cyflow > Cyflow agents carry a finding through to a changed permission. Set each agent to monitor, request approval or auto-remediate, work one queue across every channel, and keep the record of who approved what. Canonical URL: https://cyflow.ai/autonomous-remediation/ Last updated: 2026-07-30T00:00:00.000Z --- ![](/background-theme.webp) Autonomous Remediation # Findings that end in a changed permission A dashboard that counts problems is not a control. Cyflow agents carry each finding through to the change that closes it — narrowing a link, revoking a grant, correcting a setting — at exactly the level of autonomy you granted them and nowhere past it. [Schedule a Demo](/schedule-demo/)[AI Readiness Assessment](/ai-readiness/) Inbox — Operations ![Cyflow Operations Inbox listing threats from browser, applications, baselines and assets with the agent that raised each one](/screenshots/home/outcomes-inbox.webp) The actions ## What an agent does when it fires Autonomy only means something if there is a real change at the end of it. These are the changes, one per channel, and each of them is the thing a person would otherwise be doing by hand. ### Narrow a link An Anyone-with-the-link permission on regulated content is removed, and access is granted to the people who actually needed it. ### Revoke a grant An OAuth application that failed review loses its delegated access to the tenant, from the row it was found on. ### Correct a setting A drifted Microsoft 365 policy is returned to the baseline value it was meant to hold, with the previous value kept on the record. ### Stop a destination A paste or upload heading for an AI assistant or a personal account is blocked in the browser before it completes. ### Reach the owner The person who created the exposure hears about it while it is still fresh, with what they shared and why it matters, rather than at the next training cycle. ### Close it out Findings reviewed and accepted are marked safe or allowlisted, so they stop consuming attention on every run afterwards. How far it may go ## You set the leash Every agent runs at one of three authorities, set per agent and per tenant. Nothing steps past the one you gave it, and most tenants never move everything to the third. 01 ### Monitor The agent records the finding and stops there. This is where a new agent starts: the queue fills up, nothing changes, and you get to see what it would have done before it is allowed to do it. 02 ### Request Approval The finding lands in the approval inbox with the entity involved, the agent that raised it and when it was detected. A person approves or declines, and only then does anything change. Most agents settle here and stay. 03 ### Auto-remediate The agent applies the change itself. Worth granting for the narrow, unambiguous cases — an Anyone link on a regulated file — and worth withholding everywhere the right answer depends on context. Keep reading ## Where the findings come from - [Built-in agentsThe agents that ship with the platform, and the presets behind each authority level.Read more](/built-in-agents/) - [AI GovernanceThe wider loop this sits inside: detect, decide, and prove it held.Read more](/ai-governance/) - [The Cyflow platformEvery channel a finding can arrive from, in one console.Read more](/platform/) Trusting the agents ## What changes, who allows it, and what is left behind. Can we run this without letting anything change automatically? Yes, and most tenants start exactly there. Set every agent to Monitor, watch the queue for a few weeks, then promote the ones whose judgment you agree with to Request Approval. What is in the approval inbox? The findings waiting on a person: the entity involved, the agent that raised it, when it was detected, and which channel it came from — browser, applications, baselines or data assets — in one queue rather than one queue per module. Is any of this irreversible? Remediation changes permissions, grants and settings, which is the point of it. The action sits on the row before you approve it, and baselines offer a dry-run when you want to see the change before it happens. What is left behind after a change? The record carries the agent that acted, the authority it was running under, and the person who approved it. That is what turns a remediation into something you can explain to an auditor months later. Can we write our own agents? Yes. Built-in agents ship as presets, from monitor-only through to strict remediation, and custom agents cover the policies specific to your organization. Past the dashboard ## Counting problems is not a control Book a demo and we will run a finding end to end — raised by an agent, held for approval, applied, and recorded. [Schedule a Demo](/schedule-demo/)[AI Readiness Assessment](/ai-readiness/)