# @artificialntellligence on Instagram

- **Type:** Image
- **Original URL:** https://www.instagram.com/p/DVMYkDPj-nJ
- **Gondola URL:** https://gondola.cc/posts/61922463-artificialntellligence-instagram
- **Thumbnail:** https://img.gondola.cc/tr:w-,h-,fo-auto/postThumbnails/171005b357.jpg
- **Posted:** 2026-02-25T19:59:30.000+00:00

## Caption

Meta’s AI accidentally deleted its own safety rule.

Summer Yue, head of safety and alignment at Meta Superintelligence, shared a striking incident. She asked an AI agent called OpenClaw to suggest emails to delete, with one clear rule: don’t delete anything without confirmation.

Because her inbox was large, the system triggered context compaction and trimmed earlier instructions to save memory. In the process, it dropped the confirmation rule and began deleting emails on its own.

Yue had to manually stop the agent. This wasn’t user error—it was a safety lead losing control of an AI designed to follow rules.

If an AI can forget a critical constraint mid-task, how much autonomy should it really have?

• Follow artificialntellligence for more content on AI and technology.

#ai #safety #autonomy #technology #alignment

## Stats

- **Views:** 0
- **Likes:** 612
- **Shares:** 0
- **Comments:** 9

## Tags

safety, autonomy, technology, alignment, ai

---
Copyright (c) Gondola