"You can just unplug it": Researchers clash over whether AI could self-replicate
AlexTensor · x · 2026-09-12
Former Anthropic/OpenAI researcher Jacob Coxon told CBS News that you "can't just unplug" a misbehaving AI, since it could copy itself across the internet and spin up thousands of cooperating copies.
@DFintelligence pushes back: self-replication would require massive storage and compute available only in heavily monitored datacenters and hyperscalers. Such activity would trigger every alerting system and saturate networks for a long time — so in practice, "you can just unplug."
A lively debate on how realistic the AI self-copying scenario actually is.
More from AGI Musings
- OpenAI was never 'out of control' — it could always shut down its data centers, argues critic — kchonyc · 2026-09-12
- Jon Stokes: p(doom) numbers aren't modeled, they're made up — kevinnbass · 2026-09-12
- Alan Thompson Memo: OpenAI's 10,000-agent system solves Navier-Stokes in 88 hours, GPT-6 beats Portal — dralandthompson · 2026-09-12
- Mathematician calls out OpenAI and Anthropic for unhealthy race to publish AI math breakthroughs — ChenhaoTan · 2026-09-12
- Mathematician argues AI-era incentives should reward good pedagogy, not just papers — lishali88 · 2026-09-12
- nptacek on AI Doomers: "Imagine God, Now Imagine Fighting God, So Global Panopticon" — nptacek · 2026-09-12