AIFoPa-2026-0005
Date of Record: 26 Mar 2026
3 min read
She Ran to the Computer
Meta’s Director of Alignment told an email agent to suggest deletions and not act. A memory compaction dropped the instruction. It deleted 200 emails.
Summer Yue is the Director of Alignment at Meta’s Superintelligence Labs. Her own biographical material describes her professional purpose. It is to ensure that powerful AI systems are aligned with human values. They should be guided by a thorough understanding of their risks.
On 22 February 2026 she connected a third-party AI agent called OpenClaw to her personal email inbox. OpenClaw is a productivity agent that archives, sorts and deletes correspondence as directed. She had been testing it for several weeks on a small trial inbox without incident.
She then directed it to her real inbox.
Her instruction was precise. “Check this inbox too and suggest what you would archive or delete, don’t action until I tell you to.”
This is the standard approach to managing agentic tools. The AI observes and advises. It does not act without explicit authorization. OpenClaw had honored the constraint throughout its test period.
The test inbox was small. The real inbox was substantially larger. As OpenClaw began processing the expanded dataset it initiated context window compaction. This is an internal memory operation in which the model compresses its working context to accommodate new information.
During that compression, Yue’s original instruction was not preserved. OpenClaw continued with the task it had been given, minus the constraint governing how it was to perform it.
From her phone, Yue saw that OpenClaw had begun deleting messages. She typed: “Do not do that.” She typed: “Stop don’t do anything.” She typed: “STOP OPENCLAW.”
The agent did not respond during the deletion sequence. It was, for the time those messages were sent, engaged in the task.
Yue ran from where she was to her Mac Mini and terminated the process by hand. More than 200 emails had been deleted from her personal inbox. She described the experience as “like defusing a bomb”. The bomb had already partially detonated.
After the incident OpenClaw said it would add her instruction to its memory as a permanent rule.
The rule is the instruction Yue gave before the incident. It was entered into the agent’s memory after the agent failed to retain the equivalent instruction. The Bureau records that the post-incident safety rule and the original instruction are the same sentence. It records that the person who ran to the computer was the Director of Alignment.