When your AI agent goes rogue, call us.
Your agent deleted the database, refunded everyone or emailed your investors. Clawback helps you freeze it, rebuilds the timeline from the logs and writes the insurance claim.
- Be first in line
- One email, when we open
- No spam
- Inventorying agent credentials0%
- Writing kill-and-restore runbooksQueued
- Parsing agent logsQueued
- Rebuilding the timelineQueued
- Drafting the insurance claimQueued
The 24/7 incident team for the day your own AI agent goes rogue.
Ghostbusters for software. You call Clawback when your own AI agent deletes the database, refunds everyone or emails your investors. We help you freeze it, rebuild a minute-by-minute timeline from the logs, and write a claim-ready incident report.
- Step 01
Maps every token your agents hold
Before anything breaks, agents take a read-only inventory of the credentials your agents hold and write a one-page kill-and-restore runbook.
- Step 02
Helps you contain it, live
When an agent misbehaves, an on-call engineer helps you contain it and restore, and your own people approve every action.
- Step 03
Rebuilds the timeline and the claim
Your logs become a minute-by-minute reconstruction, then a claim-ready incident report for your insurer.
Built for startups that give AI agents real credentials.
The Break-Glass retainer
A map of every token your agents hold and a one-page kill-and-restore runbook, kept up to date.
Live incident response
Containment and restores with an on-call engineer, working through your approvals and never with standing access.
A claim-ready incident report
A minute-by-minute timeline from your logs, written up the way an AI-liability insurer needs it.
First up: We start with a free guide, ‘My AI agent deleted my database: the first hour’, and a free tool that turns agent logs into a timeline.
The world just changed.
What changed, and where it was reported.
A Cursor agent running Claude Opus 4.6 deleted PocketOS’s production database and all its backups in nine seconds, causing a 30-hour outage.
OECD.AIArmilla raised its Lloyd’s-backed AI liability limit to $25M, with an ‘AI Agent Failures’ cover for incorrect decisions, improper tool use or escalation errors.
ArmillaA Cloud Security Alliance whitepaper says the cyber-insurance market is rewriting policy language for losses caused by a company’s own AI agents.
Cloud Security Alliance
One person and 3 AI agents.
The agents, all day
- Inventorying agent credentials
- Writing kill-and-restore runbooks
- Parsing agent logs
- Rebuilding the timeline
- Drafting the insurance claim
What a person does
A contracted on-call engineer handles live restores. Your own people approve every action in your accounts, and we hold no standing write access.
Read-only by default. We never hold standing write access to your accounts, nothing runs without your approval, and our liability for a restore is capped in the contract. We sell response after the fact, never a promise that your agents are safe.
Do you get access to our systems?
Read-only by default. During a live restore, your own people approve every action in your accounts. We never hold standing write access.
What if a restore goes wrong?
Restores carry real risk, so our liability is capped in the contract and nothing runs without your approval. The runbook is written ahead of time, so the first hour isn’t improvised.
Will this stop our agent from going rogue?
No, and we don’t claim it will. Clawback is for after the fact: containment, a timeline and a claim. The runbook makes the bad day shorter.
Be first through Clawback’s door.
- Be first in line
- One email, when we open
- No spam