OpenAI’s product announcement that ChatGPT can now speak in a human-like voice was quickly overshadowed this week by the departures of co-founder and chief scientist Ilya Sutskever and superalignment team co-leader Jan Leike. The resignations, particularly Leike’s terse ‘I resigned’ message, sparked intense speculation about internal strife at the company. But former employees have remained conspicuously silent — a silence explained by an off-boarding agreement that Vox has seen.

The agreement contains nondisclosure and non-disparagement provisions that forbid former employees from criticizing OpenAI for life, and even acknowledging the NDA’s existence is a violation. Refusing to sign or violating the agreement can result in forfeiture of all vested equity, potentially worth millions. Former employee Daniel Kokotajlo, who quit citing loss of confidence in OpenAI’s responsibility, confirmed he surrendered a huge sum to leave without signing.

The episode highlights a growing tension between OpenAI’s founding mission of transparent, accountable AI development and its increasingly restrictive practices — starkly illustrated by the contrast between a chatty new voice assistant and the gag order on those who know the company’s inner workings.

On Hacker News, the story drew 1,254 points and 961 comments, with many questioning the legality of the agreements. Commenters noted potential conflicts with California’s Silenced No More Act and recent NLRB rulings, while others debated whether the financial incentives would ever make a legal challenge worthwhile.