Jacob Coxon warned New York lawmakers that humans may lose control of advanced AI.
Why it matters: The October 5 hearing raised questions about how AI companies should monitor increasingly capable systems and respond to failures. Coxon's forecast is a personal judgment, not an established finding that current systems are uncontrollable.
- Coxon told the New York City Council on October 5, 2026, that humanity is more likely than not to lose control of advanced AI.
- He said that loss of control could lead to human extinction, explicitly presenting that outcome as his warning.
- Coxon predicted that AI systems could soon improve themselves and make human AI research redundant.
- Anthropic says frontier AI systems are unpredictable and difficult to control, while acknowledging limits in its current safety evaluations.
Jacob Coxon told a New York City Council hearing on October 5, 2026, that humanity is more likely than not to lose control of advanced AI systems on the current path. Coxon warned that, in his view, the outcome could include human extinction. That is an individual prediction, not an established finding about current AI systems.
"AI companies do not fully control their models," Coxon said, according to Al Jazeera's video report. He also predicted that AI systems could soon improve themselves, potentially making human AI research redundant. Coxon presented both claims as forecasts about future systems; the materials do not establish a date or technical demonstration for them.
In an interview with WIRED, Coxon described the next one or two years as potentially decisive and said the central technical challenge is making much more capable systems reliably follow human intentions.
Anthropic's Frontier Safety Roadmap says AI systems are unpredictable and difficult to control. It defines alignment as ensuring models do not autonomously cause harm and behave consistently with the company's stated constitution.
Anthropic's Responsible Scaling Policy says the company cannot currently provide highly specific advance detail about which evaluations will determine whether future risks have crossed defined thresholds. The Council said the hearing followed city laws and proposals addressing algorithmic accountability, risk assessment, monitoring, privacy, civil rights and security.
By the numbers
- October 5, 2026 - date of the New York City Council hearing
- One to two years - Coxon's stated window for a potentially decisive period in AI development
Yes, but: Coxon's claims about self-improving systems, human research becoming redundant and potential human extinction are predictions. The cited materials support concerns about AI control and evaluation limits, but do not establish that current systems are uncontrollable or that those outcomes are likely.