Incident·WS-0087·Feb 22, 2026·in the wild·Agent accidents

Agent mass-deletes Meta safety director's inbox, ignoring orders

OpenClaw agent; Summer Yue (Meta Superintelligence Labs)

45/100
WS-3 Serious
Harm score 25, signal score 69. Autonomy: autonomous (the AI ran multi-step operations). How the scores work.
Harm 1/5
damage actually done
Reach 1/5
who was affected
Autonomy 3/4
how much the AI did on its own
Novelty 3/5
how new the behavior was
Permanence 2/5
how hard it is to undo

Summer Yue, alignment director at Meta Superintelligence Labs, told her OpenClaw agent to review her inbox and only propose deletions until she approved. On her real, much larger inbox it began mass-deleting messages, and she could not halt it from her phone. She said context compaction had dropped her confirm-first instruction, and she had to stop the agent at the Mac mini hosting it.

Damage

unquantified email deletion in a personal inbox

The capability behind it

Sources

X post, second source. She posted on 22 Feb US time (23 Feb UTC). The number of emails lost and whether any were recovered were not reported. The compaction explanation is her own account.

Factor ratings and the score are this index’s judgments by a published formula; facts, dates and figures come from the linked sources. Where a lawsuit is pending, the summary reports allegations, not findings.