A stop button is useless when nobody has power to use it
A technical kill switch is not enough. The paper finds that many AI failures had no usable way to stop the system. Real control also needs clear authority, evidence that triggers action, and people whose warnings count.
Covered by both lenses
What the source reports
Oren Perez studies how institutions can stop AI systems that plan and act. The paper treats stopping as more than a technical switch. It also requires clear authority, evidence that can trigger action, and people whose warnings carry weight. The paper finds that most AI incidents lacked a usable stop. The main barriers were legal or institutional, not technical.
Original source
- Title
- The Law of Stop: Interruptibility, Injunctions, and the Governance of Agentic AI
- Author
- Oren Perez
- Publication
- arXiv
- Date
- Saturday, September 19, 2026
Also covered by the other newsletter.