In the space of a single Friday, OpenAI shed three senior leaders, shut down a division it had publicly championed just months earlier, and quietly signalled that its ambitions for scientific AI are being subsumed into a coding application — all while the company accelerates towards a reported initial public offering.Three OpenAI Executives Exit on the Same DayThe departures, confirmed on 17 April 2026, read like a restructuring memo come to life.
Kevin Weil, who had served first as OpenAI's chief product officer before moving to lead its science initiative, announced his exit publicly.
Srinivas Narayanan, the company's chief technology officer for enterprise applications, told colleagues internally that he was stepping away to spend time with his family.Bill Peebles, who had been heading the Sora video-generation product, posted on X that he too was leaving.
Three senior departures in a single day would be notable at any large technology company.
At OpenAI, where executive turnover has become something closer to a recurring event, it lands as the latest chapter in a leadership story that has been unravelling for some time.Weil addressed his departure directly on social media: "Today is my last day at OpenAI, as OpenAI for Science is being decentralized into other research teams.
It's been a mind-expanding two years, from Chief Product Officer to joining the research team and starting OpenAI for Science."What Was OpenAI for Science — and Why Has It Been Dissolved?OpenAI for Science was, until recently, one of the company's more publicly visible bets on the transformative potential of frontier AI.
Weil joined OpenAI in June 2024 as chief product officer, bringing a product background from Instagram and Twitter.
By September 2025, he had pivoted internally to lead the initiative, which was designed to attract world-class academics and apply the capabilities of GPT-5 to hard problems in physics, biology and chemistry.In January 2026, the team launched Prism — a dedicated web application intended to give researchers a tailored AI workspace.
That same month, MIT Technology Review profiled the initiative as active and ambitious, drawing attention to GPT-5.2's 92 per cent score on the GPQA graduate-level science benchmark, a striking leap from GPT-4's 39 per cent.
Weil spoke at length, in that profile, about the importance of "epistemological humility" and self-fact-checking in AI models.Three months later, the division no longer exists.
Prism has been sunset, and the roughly ten-person team that built and ran it has been folded into Codex — OpenAI's AI coding application —....
