Current and former staff demand a right to warn

Thirteen current and former employees of OpenAI, Google DeepMind and Anthropic published an open letter, “A Right to Warn about Advanced Artificial Intelligence,” arguing that AI companies could not be relied on to voluntarily share information about risks with the public and that existing legal protections did not cover their employees. Seven signatories were named, including former OpenAI researchers Daniel Kokotajlo, Daniel Ziegler, William Saunders, Jacob Hilton and Carroll Wainwright, and Google DeepMind’s Ramana Kumar and Neel Nanda; six others signed anonymously, citing fear of retaliation. Turing Award winners Yoshua Bengio and Geoffrey Hinton, and AI safety researcher Stuart Russell, endorsed the letter without themselves being current or former employees of the companies concerned.

The letter argued that AI firms held strong financial incentives to avoid effective oversight and possessed non-public knowledge about their systems’ capabilities, limitations and risk-mitigation efforts, while broad confidentiality and non-disparagement agreements discouraged employees from raising concerns, and ordinary whistleblower law offered no protection because most of the risks in question were not yet illegal. It asked companies to commit to four things: not enforcing agreements that bar risk-related criticism or retaliating financially against employees who raise concerns; creating verifiably anonymous channels for reporting risks to a company’s board, to regulators and to independent bodies; supporting a culture of open criticism where trade secrets are still protected; and not retaliating against employees who go public about risks after internal channels have failed.

The letter followed weeks of reporting on OpenAI’s exit agreements, which had required departing staff to sign non-disparagement clauses or risk losing vested equity — a policy OpenAI said it had not enforced and would remove — and arrived shortly after Jan Leike’s resignation and the dissolution of OpenAI’s Superalignment team. It became a reference point in subsequent arguments over whether internal dissent at frontier labs could reach the public at all; Kokotajlo went on to co-found the AI Futures Project and co-write the AI 2027 forecast.