U.S. AI policy went in opposite directions within three days. President Trump did not sign an executive order on AI and cybersecurity scheduled for May 21 local time and put it off, and on the same day California Governor Gavin Newsom signed an executive order overhauling labor policy in response to large-scale job displacement caused by AI. The federal government released the brake on regulation, and the state government began reworking the employment safety net. Within one country, two governments dealing with AI are looking at different clocks.
The core of the postponed draft executive order was "advance access." It would give the government and critical infrastructure operators such as banks the right to test new models 90 days before AI developers release them, and it was designed around voluntary participation rather than a mandate. It reportedly did not include mandatory government approval or verification procedures for frontier models, and instead was arranged to fold AI companies into existing cybersecurity information-sharing programs. The text was prepared over a month by White House Chief of Staff Susie Wiles, the science and technology adviser and the national cyber director, among others, who took input from AI companies.
President Trump told reporters that he put off signing because he did not like certain aspects of the executive order, and because he did not want to get in the way of the U.S. advantage over competitors such as China. The signing had been prepared as an event bringing the chief executives of major AI companies to the White House. With the executives invited to Washington, the event was called off just before it was to take place.
The lines of conflict inside the White House were clear. The security and national security side held that there must be a verification mechanism before powerful models reach the world, while the pro-business, Silicon Valley camp opposed it on the grounds that such procedures themselves slow development. Former White House chief strategist Steve Bannon and conservative activist Amy Kremer, among others, have called for mandatory government verification before release.
David Sacks, co-chair of the President's Council of Advisors on Science and Technology, is said to have warned the president in a phone call that the executive order could work against the United States in its competition with China. Behind the discussion lay cybersecurity concerns surrounding Anthropic's AI system "Mythos."
Behind California's move are actual job cuts. Meta cut 8,000 jobs, 10 percent of its total workforce, citing its AI shift and expanded investment, Cisco let go of 4,000, and Amazon also cut several thousand. Anthropic CEO Dario Amodei predicted last May that half of entry-level white-collar jobs could disappear within five years. Layoff notices are arriving before employment statistics catch the change.
Governor Newsom's executive order spreads the response across several tracks. It calls for studying subsidies for companies that keep existing employees rather than replacing them with AI, and it expands job training aimed at software developers and marketing and sales workers. Expanding "work share," which cuts hours instead of laying workers off and covers part of the lost wages with unemployment benefits, expanding unemployment insurance benefits, broadening profit sharing such as severance pay and stock compensation, and expanding worker-owned companies also went on the list for review. It even includes a plan to distribute basic assets to state residents in the form of stocks, bonds or stakes in a sovereign wealth fund.
Deadlines were attached as well. Within 90 days the state government must build a dashboard that uses unemployment insurance data to track the effect of AI on employment by industry, and within 180 days it must lay out safety net policies for workers who lose jobs to AI. Governor Newsom said California has not sat and watched the future arrive and will not do so going forward, and that it is time to rethink the entire system, including how people work and how they are prepared for the future. The budget and funding sources were not included in this executive order.
The state legislature went further. On May 19, two days before the executive order, the California Senate passed the "No Robo Bosses Act," which bars firing or disciplining employees solely on the basis of AI or automated decision systems, by 29 votes to 9. It would keep algorithms from finalizing personnel decisions without a final human check, and it still awaits Assembly review. Governor Newsom vetoed a bill with a similar aim last year.
The character of the solutions each side offers also diverges. Elon Musk wrote in April that the best response to AI-driven unemployment is for the federal government to issue checks guaranteeing universal high income. That is after-the-fact compensation that hands out cash, and the work share and retraining California has put forward are attempts to hold on to the jobs themselves. Whether to top up income or to preserve work is likely to be the line of debate ahead.
Last month there was a shooting at the San Francisco home of OpenAI CEO Sam Altman, preceded by an attempted firebombing, and anti-AI protests of several hundred people have continued outside the offices of OpenAI and xAI. The president has been rolling back safety-centered policy since scrapping the previous administration's AI executive order in January 2025, and this postponement confirmed that stance once again. As a result, setting the threshold for AI development in the United States is proceeding in a form where the federal government lets go and the states take hold.
When California's dashboard opens in 90 days, how many people have left which industries will appear in visible form for the first time. From then on the debate over AI and employment will be argued with results rather than forecasts. While a signing was being put off in Washington, Sacramento started its clock, and the times the two clocks show are growing further apart.
