Daniel Kokotajlo questions AI labs' safety promises
The former OpenAI researcher tells Steven Bartlett why he left, then explains how automating AI research could accelerate the technology's development.
Why this story? 1,737,497 views so far, and here is what is inside it.
ChatGPT Whistleblower WARNS: “You’re Being LIED To"
Things we can learn from the video
Press to jump
- 0:15Kokotajlo describes forecasting dangerous AI capabilities at OpenAI and becoming disillusioned with its leadership and incentives.
- 4:43He explains his resignation, citing internal rationalizations and restrictions on publishing research.
- 7:18Refusing a non-disparagement agreement initially put his equity at risk; he says he ultimately kept it.
- 9:34Automating coding and research could let AI systems help build their own more capable successors.
- 12:11Bartlett introduces AI 2027, and Kokotajlo outlines the scenario's predicted milestones and timelines.
- 16:35Kokotajlo distinguishes AGI from superintelligence and explains why cognitive abilities could advance faster than robotics.
"I think you should judge people by their actions, not by their words," Daniel Kokotajlo tells Steven Bartlett. It is the central argument of an interview excerpt on The Diary Of A CEO Clips: the former OpenAI researcher sees a gap between AI companies' public missions and the incentives shaping their decisions.
Kokotajlo joined OpenAI to forecast AI's development and evaluate dangerous capabilities. He describes becoming disillusioned with leadership, restrictions on publishing research and what he calls "power-seeking incentives." In his account, the fear that another company will gain control of powerful AI becomes a justification for pressing ahead yourself.
The departure story gives that argument a concrete personal stake. Kokotajlo says he refused an exit agreement containing a non-disparagement clause, initially putting roughly $2 million in equity at risk. He explains that the company later reversed the forfeiture policy following public criticism and questions from employees. "I got to keep the money," he says.
The clip's title, "ChatGPT Whistleblower WARNS: "You're Being LIED To"", promises an exposé. The conversation is chiefly his account of company culture and his argument about where its incentives lead, rather than a demonstration of a particular false ChatGPT answer.
The second half turns to AI 2027, the scenario forecast he co-authored. Kokotajlo explains a potential feedback loop: companies automate coding and research, then use those systems to develop more capable AI. His timelines and bleak risk estimate are forecasts, not events that have already happened.
He also separates intelligence from physical dexterity. In his account, AI could become vastly better than humans at cognitive work before robots become equally capable in the physical world.
Published on Trari
Where next?
What viewers say
Sarah Connor is shaking her head in disbelief right now....
@moopiusSee it on YouTube"Sometimes it's good to take a stand on principle".... PREACH
@kubanski10See it on YouTubeReally brave of him. Thank you. We need the inside info.
@fabialenoci7530See it on YouTube
Public comments from the video's page on YouTube
- Channel
- The Diary Of A CEO Clips
- Subscribers
- 2,880,000 ▲ 40,000 this week
- Channel views, last 7 days
- 12,711,747
- Channel views, pace per day
- 991,715 (on a normal day: 1,704,698)
- Video published
- October 20, 2026
- Video views so far
- 1,737,497 (5 days online)
- Video views, pace per day
- 283,543 (on a normal day: 426,956)
The numbers
The excerpt has drawn about 1.7 million views in five days. Its audience is not unusually large for this channel; the substance of Kokotajlo's warning is the story.
💡 Test your knowledge
What does AGI stand for?
AGI stands for artificial general intelligence, a term for AI with broad capabilities across different intellectual tasks.


