OpenAI's exit agreements are found to claw back equity

Vox reporter Kelsey Piper published leaked OpenAI off-boarding documents showing that employees who left the company had to sign a general release containing a lifetime non-disparagement clause — and a further clause barring them from ever disclosing that the non-disparagement clause existed — within 60 days, or forfeit all vested equity they had earned, in some cases understood to be worth millions of dollars. Piper reported the language had been present in OpenAI’s standard exit paperwork since at least 2019, and separation letters seen by Vox had been signed by OpenAI’s chief operating officer as recently as late April 2024.

The report prompted a public apology. In statements the following day, chief executive Sam Altman said he had not been aware the provision existed, calling it “one of the few times I’ve been genuinely embarrassed running OpenAI,” and said the company was rewriting its exit paperwork and would release any already-departed employee from the clause on request. OpenAI subsequently confirmed it had removed the non-disparagement and clawback language from its standard agreements.

The story landed during a week of wider turbulence at the company: chief scientist Ilya Sutskever and Superalignment co-lead Jan Leike had both resigned days earlier, and criticism of how OpenAI’s November 2023 board crisis had been handled was still circulating. Coming so soon after departures explicitly framed around safety disagreements, the clawback provision read to critics as a structural reason those disagreements had stayed private — anyone leaving over a safety concern had a strong financial incentive to say nothing about it. The episode became one of several moments that year in which OpenAI’s internal dissent surfaced only after it had already left the building.