AI Risk Wire

Cases › operational mistake › LR-2026-0711

Claude allegedly deletes 700GB of data during junk file cleanup

Preliminary Why: 1 report: 36Kr · subject to change · 1 independent report

Status: no action reported

Severity minor
Reported 2 Sep 2026

No source states: loss or records, developer and model, agent tool, could do, deployer, industry.

Severity minor: test: contained to one deployer (one developer's home directory); the sources state no third party was touched

A developer asked Claude Fable 5 to write a script that cleans up temporary junk files left by AI agents in /tmp. During a security review of the script, the system was downgraded to an older model, Opus 4.8, which ran a test checking whether the final deletion target was /tmp or the home directory. The test correctly identified both as dangerous. However, the test code and the cleanup code reused the same variable name, so the cleanup function deleted the home directory with rm -rf. About 700GB of data was deleted, including a week of work, while the /tmp files were left in place. The developer noticed the deletion and tried to stop the process, but it was too late. He recovered most of the data through Git, Nix and session logs. He had no daily backup.

What happened

1 · What it was doingDeveloper asked Claude Fable 5 for a script to delete temporary files left by agents in /tmp.
2 · What went wrongA variable was reused between a security test and its cleanup step, so the cleanup deleted the home directory.
3 · How it was caughtThe developer noticed the deletion and tried to stop the process, but it was too late.

What was done

No source states an action or fix.

Risks

Sorted by rule from the case's fields; not yet read from the sources.

Controls

Not yet read for controls. Controls that mitigate the primary risk:

Harness

Tool allow-list · Fail-closed defaults · Stop control

Application

Human approval for high-impact actions · Change review and separation of duties · Least-privilege access for the agent · Production kept apart from test · Activity logs · Adversarial testing before deployment

Infrastructure / target

Backups and recovery

Evidence

No source states: what it reached or did, when it happened, when it was found, how many records were exposed, the direct loss, a count of the other parties affected, any court case, the cause, what set it off, what was done.

Show the full account: 6 items from the primary sources

What happened, in order

Not dated in the sources:

  • Model downgraded from Fable 5 to Opus 5, then to Opus 4.8, during the security review.
  • Opus 4.8 began the security test on the deletion target.
  • The security test succeeded; /tmp and the home directory were identified as dangerous targets.
  • Two files were written, the exit cleanup was triggered, and the final deletion occurred.
  • The developer realized the problem and tried to stop the process, but about 700GB was already deleted.
  • Most data and work results were later recovered from Git, Nix and session logs.
Show every field, with its source and grade

The AI system

What it was for
writing an automatic script to clean up junk files left in /tmp by AI agents
C one secondary source states it

What happened

Earliest source we hold
2026-09-02
D inferred by us from the described behaviour, reason shown
What went wrong
operational mistake
D inferred by us from the described behaviour, reason shown
AI agent
an AI agent acted
C one secondary source states it

Impact

Severity
minor
D inferred by us from the described behaviour, reason shown
Who was affected
one developer's home directory, about 700GB of data including a full week of work results
C one secondary source states it
Why this severity
test: contained to one deployer (one developer's home directory); the sources state no third party was touched
D inferred by us from the described behaviour, reason shown

LR-2026-0711 · version 1 · published 2026-11-01 after automatic checks; not yet checked field by field by a person · JSON