The U.S. federal government announced its AI safety measures as a voluntary agreement with six large technology companies. There were no new regulations. In California, Governor Gavin Newsom signed SB 947, the so-called 'No Robo Bosses Act,' which bars employers from firing or disciplining workers using artificial intelligence alone. Washington relied on corporate promises, and California brought the law to bear at the moment a company lets a worker go.
A voluntary agreement is an approach in which companies promise to comply on their own instead of being compelled by law. The Trump administration's agreement includes strengthening companies' internal controls and undergoing independent audits. An independent audit is a procedure in which a third party from outside the company comes in to check whether safeguards actually work. The announcement did not reveal who will conduct the audits or when.
Critics pointed out that there is no enforcement mechanism to hold the six companies to account if they break the agreement.
President Trump recently promoted the administration's AI policy by showcasing the federal government's new AI chatbot, and also hosted a dinner with Big Tech executives at the White House. Treasury Secretary Scott Bessent likened AI company executives who call for government regulation to Hannibal Lecter, the serial killer from the movies. White House officials took the position that AI development must keep moving forward even as concerns are raised that jobs could decline.
The California law differs in its target from the outset. The law sets out what employers must not do when they use AI in the workplace to let people go. The 'firing decided by AI alone' that it blocks refers to cases in which a dismissal or disciplinary action is finalized based solely on a program's evaluation, without human review.
SB 947 prohibits such firings and disciplinary actions. Given that Governor Newsom vetoed a 'robot boss ban' bill with the same intent last year, he has changed his stance within a year.
The law's effects will not become apparent as soon as it is signed. Whether a human must make the final decision, and whether workers have a way to be notified of the outcome and challenge it, will take shape only once implementing rules and the first disputes emerge. Also worth watching is how pro forma reviews, in which an employer passes an AI evaluation through by adding just a manager's signature, will be screened out.
Employment outlooks are mixed. The baseline scenario of one forecast projected that AI and industrial restructuring would create about 41 million new jobs, and that labor equivalent to about 36 million jobs would be affected by automation.
The federal agreement is aimed at the six companies that build AI models. The California law is aimed at employers that buy and use those models. As the target of regulation shifts from developers to users, the unit of control also narrows from 'Is this model safe?' to 'Did a person take responsibility for this termination notice?'
For people working in California, what changes is who finalizes firings and disciplinary actions. AI will continue to analyze work records and assign scores, but those scores alone cannot finalize a firing or disciplinary action. The federal agreement will be tested by whether the six companies keep their promises, and the state law by employers' actual personnel decisions.
