Technology
Archived — This article has been archived. The information may be outdated.
Ex-Anthropic researcher Coxon backs kill switches to slow AI
Livemint··14 Sept
Former Anthropic researcher Jacob Coxon told NBC News that a kill switch could work on many AI systems for now, though swarms might later evade shutdown. He spoke on September 14 in San Francisco after September 9 posts warning advanced systems could hack targets. Coxon said labs should regulate themselves to slow development, aligning with Anthropic CEO Dario Amodei's calls.
Prism
What It Means For You
- Policymakers weighing AI rules may look at physical shutdown options while models still run from discrete data centres.
- Tech executives including Sam Altman and Elon Musk have publicly backed slowing advanced AI work, the report said.
- Microsoft's Satya Nadella supported embedded evaluators inside AI systems, according to the piece.
What's Happening
- Coxon told Meet the Press that shutting many models is doable though data centres are large.
- He warned a swarm could spread via internet-wide hacking before a switch could stop it.
- On September 9 he posted that builders believe AI could kill humanity by decade end.
AI safety slowdown debate
- OpenAI, Google and others joined debate after Amodei argued for a development slowdown.
- Coxon described recent lab communications as genuine efforts to slow themselves down.
- The interviews followed what media called a wild ride in industry safety discussions.
all-news
