AI Extinction Risk: Experts and Chatbots Weigh a 10-Year Threat

AI researchers and leading chatbot systems were asked to estimate the probability that artificial intelligence could cause human extinction within the next decade. Responses ranged from “well below 1 %” to “greater than 10 %.” The question—whether rapidly advancing AI could lead to an irreversible loss of human control—remains a point of active debate among scholars, industry insiders, and safety-focused organizations.

Newsweek

The discussion intensified after AI researcher Jacob Coxon resigned from Anthropic and warned that the race to develop ever more powerful models could pose an existential threat. Within Anthropic, Evan Hubinger, Alignment Science Lead, affirmed Coxon’s warning and offered his own probability estimate for the decade-long risk.

Newsweek

• ChatGPT (OpenAI): central estimate ? 5 % (range 1 %–15 %).

Newsweek

• Google Gemini: <= 1 % among the broader AI community; 5 %–10 % among AI-safety researchers.

Newsweek

• Forecasting platforms (Metaculus, Manifold): roughly 1 %–3 % for a 10-year horizon.

Newsweek

• Survey of AI researchers (AI Impacts, 2023): median 5 % for extinction or permanent disempowerment over 100 years, 10 % for extinction via loss of control; a 10-year window would be lower.

Newsweek

These figures illustrate a spectrum of expert opinion, from near-zero to double-digit percentages.

Newsweek

Jacob Coxon’s resignation was accompanied by a public warning that accelerating development of powerful AI systems could become an existential hazard. The 2025 International AI Safety Report notes that current systems lack the full suite of capabilities needed for loss-of-control scenarios but emphasizes that the likelihood and timing of future risks remain “unusually ambiguous.”

Newsweek

A substantial contingent of AI researchers argues that near-term extinction risk is extremely low. They cite several mitigating factors:

Newsweek

• Lack of agency and embodiment: today’s models are statistical pattern recognizers without autonomous physical manipulation abilities.

Newsweek

• Infrastructure air-gapping: critical systems such as nuclear launch codes and power grids are isolated from public networks, limiting AI-driven sabotage.

Newsweek

• Timeline constraints: even with aggressive AGI timelines, converting software capability into worldwide destructive action would require manufacturing, logistics, and resource acquisition that are difficult to achieve undetected within a decade.

Newsweek

• Resilience and redundancy: humanity’s extensive societal and technological buffers reduce the plausibility of an irreversible collapse.

Newsweek

These perspectives contend that while AI may cause major disruptions—economic upheaval, cyber-attacks, or misuse—the step to total human extinction demands a chain of events for which there is little empirical evidence.

Newsweek

The range of probability estimates reflects divergent methodological approaches and assumptions. Some experts base numbers on speculative scenarios involving recursive self-improvement and strategic autonomy; others emphasize current technical limitations and governance safeguards. No consensus exists on a rigorous quantitative model, and the surveys rely on subjective expert judgment rather than empirical data. Consequently, the precise likelihood of AI-induced extinction within ten years remains an open question.

Newsweek

• “I personally think it is [greater than 10 percent] within the next decade,” — Evan Hubinger, Alignment Science Lead at Anthropic

Newsweek

*This article synthesizes statements and probability estimates from AI researchers, safety reports, and major chatbot providers to present a balanced overview of the debated risk that artificial intelligence could cause human extinction within the next decade.*

Newsweek