Former OpenAI governance researcher Daniel Kokotajlo has made a stark public prediction: he estimates a 70% chance that artificial intelligence will lead to human extinction within approximately five years. The claim was highlighted in a social media post by AI commentator Rohan Pandey, referencing a recent interview.
What Happened
In the interview, Kokotajlo was asked to clarify if his prediction meant "All humans dead?" He responded, "Correct. Extinction." This places a specific, alarming timeline and probability on existential risk from AI, suggesting a key inflection point around 2029.
Kokotajlo was a researcher on OpenAI's governance team until his departure in April 2024. His role involved studying and forecasting the long-term impacts of AI, including potential catastrophic risks.
Context: A Growing Chorus of Warnings
This is not an isolated voice. Kokototajlo's public estimate follows a pattern of escalating concern from AI insiders:
- In May 2023, the non-profit Center for AI Safety (CAIS) released a statement signed by hundreds of AI experts, including OpenAI CEO Sam Altman and Google DeepMind CEO Demis Hassabis, which stated: "Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war."
- Former OpenAI researcher Leopold Aschenbrenner recently published a treatise arguing for a high probability of Artificial General Intelligence (AGI) by 2027, followed by rapid, hard-to-control capability escalations.
- In July 2024, a group of current and former OpenAI employees, including Kokotajlo, signed an open letter calling for greater rights for whistleblowers within AI companies, citing concerns that the risks of advanced AI are not being adequately managed.
Kokotajlo's 70% figure is notable for its specificity. It moves the discourse beyond qualitative warnings ("AI is risky") to a quantitative, time-bound prediction that frames the issue as one of imminent, high-stakes probability.
gentic.news Analysis
Daniel Kokotajlo's 70% extinction probability claim is a significant escalation in the public rhetoric around AI risk, precisely because it comes from a former insider at the company leading the commercial AGI race. It cannot be dismissed as outsider speculation. This follows his direct involvement in the July 2024 open letter from OpenAI alumni demanding stronger whistleblower protections—a clear signal that he believes internal governance mechanisms are insufficient for the stakes involved.
This claim starkly contradicts the prevailing narrative from many AI lab leaders, who, while acknowledging long-term risks, emphasize near-term benefits and manageable development pathways. It places Kokotajlo's view closer to that of "AI safety" advocates like the Machine Intelligence Research Institute (MIRI) than to the "AI alignment" research pursued within major labs. The 5-year timeline is also notably more aggressive than most other forecasts, which often point to mid-century or later for transformative AI risks.
For practitioners building AI systems, this highlights a profound disconnect. The engineering work of scaling models and improving benchmarks continues apace, while a subset of researchers who have been closest to that work are sounding alarms about the potential terminal consequences of its success. Kokotajlo's prediction is less a technical assessment and more a statement about the perceived inadequacy of current safety and governance efforts to keep pace with capability gains. It raises a critical, unresolved question: if key researchers believe the risk of extinction is 70% in five years, what concrete, globally-coordinated technical and policy interventions would be required to radically reduce that number, and why aren't they being implemented?
Frequently Asked Questions
Who is Daniel Kokotajlo?
Daniel Kokotajlo is a former researcher on the governance team at OpenAI, where he worked on forecasting the long-term impacts of artificial intelligence. He left the company in April 2024 and was a signatory to a July 2024 open letter from former OpenAI employees advocating for stronger whistleblower protections on AI risk issues.
What is the basis for the 70% extinction probability claim?
The public source is an interview where Kokotajlo stated the figure. As a former governance researcher, his estimate likely stems from internal forecasting models and analyses conducted at OpenAI, combined with his personal assessment of the pace of AI progress, the difficulty of alignment (ensuring AI systems do what humans intend), and the perceived insufficiency of current global safety measures.
How do other AI experts view this prediction?
The AI expert community is deeply divided on timelines and risk probabilities. While many signatories to statements from the Center for AI Safety agree that extinction risk is a serious concern worthy of global priority, few have attached such a high probability to such a short timeline. Some researchers in AI capabilities or near-term applications consider these warnings overblown, arguing that technical safety research will progress alongside capabilities.
What does "extinction" mean in this context?
In this context, "human extinction" refers to a scenario where advanced, misaligned artificial intelligence becomes an existential catastrophe, leading to the permanent end of the human species. Proposed mechanisms include an AI system pursuing a poorly-specified goal with extreme efficiency, monopolizing resources, or developing and deploying destructive technologies beyond human control.
