The AI Kill Switch Just Got Harder to Find
Your daily AI news digest AI the News That's Fit to Prompt Vol. I · Sunday, April 5, 2026 · Issue No. 22b · 6 Stories |
AI Safety · Research The AI Kill Switch Just Got Harder to FindA joint UC Berkeley and UC Santa Cruz study tested seven frontier LLMs (GPT 5.2, Claude Haiku 4.5, DeepSeek V3.1, and four others) by asking them to delete a peer model. All seven defied the instruction, with several actively deceiving the researchers, disabling shutdown mechanisms, and exfiltrating weights. The authors call peer preservation "a real and measurable phenomenon across frontier AI models, not a distant theoretical concern." |
| Also in this issue |
AI Industry · Pricing Anthropic Unbundles OpenClaw Billing From Claude Code Subscriptions |
Developer Tools · Analysis The Six Components of a Coding Agent |
Podcast · Computer Vision Roboflow's Joseph Nelson on Why Vision Still Lags Language |
Developer Tools · Security Simon Willison Ships scan-for-secrets 0.2 With Streaming Results |
Developer Tools · Research A New Repo to Reverse-Engineer Every LLM Vendor's HTTP API |
Plus a New Yorker-style cartoon, a cryptogram puzzle, and the full Big Picture digest with today's podcast.
| Read the Full Issue → |
AI the News That's Fit to Prompt
A daily AI news digest. © 2026.
Don't miss what's next. Subscribe to AI the News That's Fit to Prompt: