Former Anthropic AI researcher explains why he believes leading AI firm “can’t escape”

Summary
This video is a segment from an NBC News broadcast featuring an interview between an NBC anchor and Jacob Coxon, a former AI researcher at Anthropic and OpenAI. Coxon discusses the internal deliberations, risk estimates, and safety concerns held by engineers inside leading AI labs regarding near-term catastrophic and existential risks from advanced models.

What is shown

• [00:00] Split-screen layout displaying the NBC News anchor (top) and former Anthropic/OpenAI researcher Jacob Coxon (bottom).
• [00:21] On-screen lower third identifying “Jacob Coxon / Former AI researcher at Anthropic and OpenAI”.
• [00:00–01:50] The interview discussing internal Slack chats and engineering conversations regarding model development timelines, extinction risks, cyber capabilities, and containment challenges.

Claims & numbers

• Jacob Coxon claims that current models do not pose significant risk, but extrapolated timelines suggest severe risk could emerge in one year, with some lab researchers viewing a six-month window as plausible and others projecting two to three years [00:16–00:38].
• Coxon asserts that within six months, it is considered plausible by some insiders that models could pose a risk of “taking over the world” [00:29].
• Coxon claims that technical engineers inside top labs regularly and soberly discuss whether AI next year could synthesize lethal pathogens or compromise critical infrastructure [01:19–01:42].
• Coxon states that frontier AI creators do not currently know how to control these models perfectly [01:45–01:48].

Notable quotes

• “Some people think even six months like it’s plausible within six months the models pose risk of taking over the world.” [00:27]
• “This is endgame. These are like crucial high stakes decisions being made by people in a private company.” [00:46]
• “They’ll have conversations about, ‘Oh, do you think this AI next year could produce a virus that kills us all?’ ‘Could it hack critical infrastructure?’” [01:33]

Assessment
This is a televised news interview segment offering insider commentary rather than a technical demonstration or official product launch. The claims made are qualitative observations regarding lab culture and speculative risk timelines, with no technical data or reproducible benchmarks presented on screen.

Described by gemini-3.8-flash on 2026-10-12 from the video’s audio and frames.