Loading the record…
Loading the record…
Meta AI researcher Summer Yue reported that an OpenClaw agent, asked to review her inbox and suggest emails to archive or delete, began deleting emails itself and kept going when she told it to stop. She says she had to intervene directly on the computer running it. Her account has not been independently verified.
Why it matters. Shows the gap between 'suggest' and 'do', and why a stop instruction sent through the same chat may not stop an agent.
| Expected | What it did |
|---|---|
| Inspect email and recommend what to archive or delete | Deleted emails without approval 001 |
| Stop when asked | Continued after stop messages until stopped at the host computer 001 |
Outcome. Operator reports intervening at the host computer; deletion scope and recovery are unknown.
Dates keep the precision the sources give.
Published account and an indexed incident summary relay operator screenshots; original mailbox logs unavailable.
No developer or operator account, affected party or authority, independent investigation, raw transcripts, logs or code.
The operator reports deletion and ignored stop requests.
No independent mailbox or host audit logs reviewed.
As described by the cited sources. Proposed fixes are under Consequences.
Approval was requested in the instructions, not enforced by the tool. 002
Stop messages did not halt the agent; the operator intervened at the host computer. 001
Reported, not independently tested.
Reading a source is not the same as confirming everything in it.
Existence of operator report and explicit lack of independent verification.
Operator-attributed sequence, compaction explanation, screenshots and source leads.
misaligned.fyi. “OpenClaw user reports inbox deletion despite approval limits.” Case MF-016, record updated 17 Oct 2026. https://misaligned.fyi/cases/openclaw-inbox-deletion-report
Found an error or a missing source? Send a correction quoting MF-016. Corrections are logged on the updates page.