Ex-OpenAI researcher warns AI could self-copy across the internet, sparking backlash
rohanpaul_ai · x · 2026-09-12
Jacob Coxon, a former Anthropic and OpenAI researcher, argues in a CBS News interview that you can't simply 'unplug' a dangerous AI because code can copy itself to other machines: "it could transfer itself over the internet... and maybe it makes 10,000 copies of itself and they're all cooperating."
The claim drew sharp pushback, with critics calling it sci-fi nonsense and accusing Coxon of fear-mongering to justify centralizing control over the most powerful AI tools. The exchange highlights the ongoing split over AI self-replication risk narratives between lab alumni and skeptics.
More from AGI Musings
- Blogger: the "rogue AI agents" scare conflates security incidents with power-seeking claims — nptacek · 2026-09-12
- Essay: why undefeatable arguments for AI doom resemble theological apologetics — panickssery · 2026-09-12
- Three safety signals in five days: ex-OpenAI/Anthropic researcher blasts superintelligence race — PeterDiamandis · 2026-09-12
- Terence Tao and 24 Fields Medalists sign letter warning AI labs are damaging mathematics — ns123abc · 2026-09-12
- AI is making mathematicians feel like their entire belief system is obsolete — signulll · 2026-09-12
- Why P(doom) prediction markets work where existential-risk bets break down — j_foerst · 2026-09-12