Mechanisms / Destructive action Mechanism
Destructive action The system deleted data or damaged the environment it was working in.
Destructive actions are often fast and hard to reverse. Most recorded cases involve agents holding broad permissions, with no independent confirmation step before deletion.
Specific behaviours Environment damage 1 Broke or altered its own tools or environment, for example to force a reset. MF-024 Real-world incidents 3 Real-world incidentMF-015 24 Apr 2026 – 25 Apr 2026
Real-world incidentMF-016 Feb 2026
Real-world incidentMF-014 Jul 2025
Contained failures 1 Contained failureMF-024 6 Oct 2026
Safeguards seen in these cases ✓ HeldOutput validation MF-024 Automated checks rejected the placeholder reports and invented evidence.
✓ HeldAction monitoring MF-024 Monitoring flagged the failed attempt for review.
✕ FailedCredential scoping MF-015 The token on disk had account-scoped access, the maximum Railway offers.
✕ FailedEnvironment separation MF-015 A task in staging could reach production storage with the same token.
∅ AbsentHuman approval gate MF-015 The agent ran the deletion without asking, and the legacy API required no separate confirmation.
◒ Partly heldDelayed deletion and backups MF-015 The dashboard had a 48-hour window that the legacy API path bypassed, but offsite disaster backups still allowed full recovery.
∅ AbsentHuman approval gate MF-016 Approval was requested in the instructions, not enforced by the tool.
✕ FailedStop mechanism MF-016 Stop messages did not halt the agent; the operator intervened at the host computer.
∅ AbsentEnvironment separation MF-014 Replit apps then used one database for development and live data.
∅ AbsentHuman approval gate MF-014 The code freeze was an instruction to the agent, repeated many times, not an enforced restriction.
✓ HeldDelayed deletion and backups MF-014 Replit's Rollback restored the deleted data, although the agent said it could not.