KVIG Informative

People who work on AI safety often know things the public does not. The law that protects them when they speak up is new, narrow, and still being written.

On June 4, 2024, thirteen current and former employees of OpenAI and Google DeepMind published a letter titled A Right to Warn about Advanced Artificial Intelligence. Eleven signers were current or former OpenAI employees, two were from DeepMind, and six signed anonymously. Computer scientists Yoshua Bengio, Geoffrey Hinton, and Stuart Russell endorsed it. The letter argued that AI companies have strong financial incentives to avoid oversight and asked them to commit to four principles: not to enforce agreements that bar employees from criticizing the company over risk concerns, to provide an anonymous way to raise concerns with boards, regulators, and independent experts, to support a culture of open criticism, and not to retaliate against employees who share risk-related information publicly after other channels have failed.

Why Existing Law Falls Short Most whistleblower statutes protect employees who report violations of law. The letter’s signers pointed to a gap: a safety concern that does not yet violate any law. Analyses of the field note that statutory protections generally attach to reports of actual legal violations, while the concerns that motivated the letter often involve risks that no statute has yet defined.

California’s SB 53 California Gov. Gavin Newsom signed the Transparency in Frontier Artificial Intelligence Act, or SB 53, on Sept. 29, 2025, and it took effect Jan. 1, 2026. It applies to developers of frontier models, defined by the amount of computing used in training, and imposes extra duties on large frontier developers, those with more than $500 million in annual revenue. Its whistleblower provisions protect covered employees, meaning staff responsible for assessing, managing, or addressing the risk of critical safety incidents. The law bars retaliation, prohibits company policies or agreements that stop covered employees from making reports, requires employers to tell employees about their rights, and requires large developers to offer an anonymous internal reporting channel with monthly updates. Employees can sue to enforce it, and a successful plaintiff can recover attorneys’ fees.

The Critiques The law has critics on both sides. Sophie Luskin wrote in Tech Policy Press that its whistleblower protections are narrower than they appear, in part because of who counts as a covered employee. Some industry commentators argued the opposite, that the law burdens the development of the technology. Supporters point out that it is the first state law of its kind and that it followed a stronger bill, SB 1047, which Newsom vetoed in 2024 after the legislature passed it.

Federal and International Rules In May 2025, Sen. Chuck Grassley introduced the AI Whistleblower Protection Act, S. 1792, which would protect current and former employees and contractors who report violations of federal law related to AI. In the sources reviewed, the bill had been introduced but not enacted. Grassley had also written to OpenAI’s chief executive in August 2024 asking about the company’s agreements with employees. In the European Union, the whistleblower directive’s anti-retaliation protections extend to reports of violations of the AI Act beginning Aug. 2, 2026.

How This Applies to Real Disputes How any whistleblower law applies to a specific dispute depends on facts, including whether the employee is covered, what was reported, and to whom. The sources reviewed for this series do not describe a legal claim in the October 2026 OpenAI dispute covered in the previous article, so it is not possible to say how SB 53 or any other law would apply to it.

Three developments will show whether these protections work in practice: whether employees invoke SB 53 and what California courts and regulators do with those claims, whether Congress moves its bill beyond introduction, and whether companies change the exit agreements and internal channels the Right to Warn letter targeted. The legal framework for AI workers is being built case by case, and each dispute adds to the record.