Summary
On April 17, 2026, OpenAI's former Chief Product Officer Kevin Weil and Sora head Bill Peebles announced they were leaving the company; enterprise CTO Srinivas Narayanan disclosed his departure the same day. All three exits occurred as OpenAI formally wound down its two highest-profile consumer research initiatives — the OpenAI for Science platform Weil had built, and the Sora video generation team — folding their personnel into other divisions. The departures were framed in internal communications and press coverage as OpenAI eliminating "side quests" in advance of an anticipated public offering.
What Happened
The departures were first reported by The Information on April 17, 2026, confirmed that same day by TechCrunch and Bloomberg from independent sources. Kevin Weil posted to X announcing his exit, writing in part: "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."
Weil had joined OpenAI in 2024 as its first Chief Product Officer, taking responsibility for consumer and developer-facing products including ChatGPT, the API, and developer tooling. In mid-2025 he transitioned to founding OpenAI for Science, an initiative announced publicly in October 2025 that aimed to build AI-powered infrastructure for accelerating scientific discovery. OpenAI for Science's first public output was Prism, a platform for automated research workflows, and its final output was GPT-Rosalind — OpenAI's first purpose-built life sciences model, announced on April 16, 2026, the day before Weil's departure.
Bill Peebles was the principal researcher behind Sora, the text-to-video model that had drawn significant public attention when first demonstrated in February 2024. OpenAI had announced on March 24, 2026 that Sora's consumer applications would be discontinued on April 26, with the API following on September 24. At its peak, Sora had approximately one million monthly active users before declining below 500,000. Compute costs to operate the platform were reported in press coverage as approximately $1 million per day.
Srinivas Narayanan, OpenAI's Chief Technology Officer for Enterprise Applications, also disclosed his departure on April 17. Narayanan had overseen OpenAI's commercial and enterprise deployment stack. His departure was reported by multiple outlets but announced via internal communications rather than a public post; no personal primary announcement was identified. TechCrunch, Bloomberg, and other outlets covered his exit in the same reporting cycle as Weil's and Peebles's.
OpenAI issued no formal press statement covering any of the three departures. The organization described the changes in terms of absorption — OpenAI for Science's work was "being decentralized into other research teams" per Weil's own post, and the Sora team's remaining personnel were redistributed across product groups.
Why It Matters
The April 2026 triple departure extended a documented pattern of OpenAI executive turnover that began with the board crisis of November 2023. The departures from 2024 to 2026 — including Co-founder and President Greg Brockman (September 2024), CTO Mira Murati (September 2024), Chief Scientist Ilya Sutskever (May 2024), and multiple senior research and product leaders — had by April 2026 replaced most of the original leadership composition. The cumulative result is a company whose executive team is largely reconstituted around enterprise revenue and pre-IPO operational execution, distinct from the research-forward composition that characterized the company through 2023.
The phrase "side quests" — circulating in internal OpenAI communications and surfaced in press coverage — carries specific content. Sora and OpenAI for Science represented different categories of organizational ambition: Sora was a multimodal generative capability demonstration aimed at establishing OpenAI's position in AI video; OpenAI for Science was a stated public-benefit commitment to scientific acceleration. Their simultaneous discontinuation in the same restructuring round documents that the resource trade-offs created by IPO preparation were real, not merely rhetorical. Codex — covering computer use, agentic developer workflows, and enterprise software integrations — was the explicit strategic priority replacing both.
The compressed lifecycle of OpenAI for Science is a notable data point for the record: the initiative was publicly announced in October 2025, received a named VP, launched its first product (GPT-Rosalind) on April 16, 2026, and was dissolved the following day. This six-month arc from public announcement to dissolution offers a concrete example of how AI lab organizational commitments respond to commercial and pre-IPO pressures. The full financial reasoning behind the decision — whether Sora and OpenAI for Science were net-negative on the margin, what internal revenue projections said, and how the IPO valuation model drove the restructuring — has not been disclosed.
§ How to read the metadata
- Landmark
- Fundamentally alters the trajectory; 2–5 per year.
- Major
- Meaningfully shifts the landscape; 2–4 per month.
- Notable
- Worth documenting; significance can be upgraded later.
- Confidence
- High = primary sources corroborate. Medium = credible secondary only. Low = provisional. Disputed = credible sources disagree.
- Contestation
- Uncontested = no formal challenge. Contested = at least one challenge open. Superseded = replaced by a later entry. Unresolved = dispute still open.