House Republican leaders are pressing ahead on a proposal to preempt state-level AI regulation, two GOP aides told Maria and Ashley last night.

Why it matters: President Trump's draft executive order to thwart state AI laws may not be ready for prime time, so his Hill allies are scrambling to get the task done via Congress.

State of play: The executive order, which was expected as soon as today, is likely to be delayed, multiple sources told Axios, and Capitol Hill is feeling the heat to provide a legislative play in the National Defense Authorization Act.

Driving the news: A flurry of proposals were circulating in Washington yesterday, from language mirroring California's major new AI law to a limited preemption model, multiple sources told Axios, noting all ideas were in flux.

Reality check: The effort is a long shot.

• The leaders of the Armed Services committees in each chamber are wary of including these provisions in the must-pass defense bill.
• Democrats, civil society groups, and conservatives who are vocally in support of states' rights, including Govs. Ron DeSantis and Sarah Huckabee Sanders, will be a tough sell.
• Republicans protective of their states' internet grants resisted tying them to AI regulation when Sen. Ted Cruz (R-Texas) tried that route this summer, and now that idea has resurfaced in Trump's draft executive order.

What they're saying: The draft executive order to preempt state-level AI regulation has been widely denounced as a Big Tech power grab by many groups, but some tech, AI and startup leaders applauded it.

• "This is obvious overreach. Congress explicitly rejected AI law preemption attached to broadband funding in a 99-1 vote," tech advocacy group Americans for Responsible Innovation president Brad Carson said.
• California State Sen. Scott Wiener, who sponsored a major AI bill that was signed into law this year, said in a statement to Axios that "Trump thinks he's the king, but he's not. He has no power to issue a royal edict canceling state laws."
• "AI has so much promise to improve people's lives — cure diseases, help us fight climate change, etc. — and as with any new technology, there are also risks. California has the power and the responsibility to both promote AI innovation and reduce those risks," Wiener added.

Sens. Ed Markey (D-Mass.) and Elizabeth Warren (D-Mass.) sent a letter to Senate Democrats yesterday calling for them to stay united in opposing the National Defense Authorization Act if it includes an AI moratorium.

• "Democrats have direct control over whether the AI moratorium is enacted into law. Our message should be simple: Congress should not give Big Tech a multi-year holiday from state oversight," the letter states.

What's next: Both chambers are aiming to vote on the annual defense policy bill in December, and lawmakers are now waiting to see the final version.

The Internal Revenue Service is rolling out Salesforce's AI agent program, Agentforce, in multiple divisions across the agency, per an announcement shared exclusively with Ashley.

Why it matters: At a time when the IRS has seen a massive reduction in its workforce due to Trump administration layoffs, it's also using AI agents for the first time.

• The IRS workforce has gone down 25% this year, from 100,000 to about 75,000 employees.

Driving the news: The IRS will use Agentforce across the Office of Chief Counsel, Taxpayer Advocate Services and the Office of Appeals, said Paul Tatum, executive vice president of global public sector solutions at Salesforce.

• After working with the IRS for a few years to modernize some basic platform technology in those departments, the agency is now bringing in Agentforce to "augment and supplement the work of these departments," Tatum said.
• Agentforce will be used for tasks like case summarization and search to try to complete customer cases more quickly.

The big picture: Government agencies are increasingly incorporating AI into their workflows, which is a massive boon for companies looking to expand their customer bases and provide examples of large-scale AI use to the private sector.

What they're saying: "Salesforce doesn't advocate for a blind AI processing tax returns without a human being involved in reviewing and supplementing it," Tatum said.

• "When the agents are built, there's a lot of guardrails put in ... [they're not] allowed to make final decisions, they're not allowed to disperse funds."
• "It's an educational process for the public and the government on the safety and trust of the agentic technologies that we're so excited about," he said.
• What the IRS decides to do with its staffing levels is up to the agency, Tatum said, but the aim of the tech is to help overworked IRS agents get through customer requests more quickly and efficiently.

Rob Fitzpatrick, a senior level counsel (technology) in the IRS Office of Chief Counsel, told Axios that it felt like the right time to start using this technology at the agency.

• In 2023, the IRS started modernizing its decades-old systems, Fitzpatrick said, and he initially resisted using AI.
• But he said he feels "it would be negligence if I didn't start now using those AI tools to take our automation and now go head-to-head with some of the law firms."

The bottom line: Fitzpatrick, who's been at the agency for 38 years, cast these changes as inevitable, and said layoffs at the IRS were likely due to factors beyond embracing AI.

• "I think all of us have to realize that the change is coming," Fitzpatrick said. "You either have to adopt the change and make yourself more efficient so that you can produce more work, or you don't, and you leave."

Humanlike AI companions are dangerous by design and should not be used by minors at all, per policy recommendations that are being floated by a collection of children's safety, family and consumer groups and were shared exclusively with Ashley.

Why it matters: There's long been a push to get kids off social media by people who say it has damaging effects on young minds, and now that fight is moving to combat chatbots.

Driving the news: The policy recommendations come from a diverse group across the political spectrum.

• It includes author Jonathan Haidt's group The Anxious Generation, along with American Compass, Fairplay, All Girls Allowed, Young People's Alliance, and the Institute for Families and Technology.
• Public Citizen, Common Sense Media, Family Policy Alliance and Consumer Federation of America are also part of the group.

What they're saying: AI companions are a "crutch" for kids that "can be entertaining, and can temporarily fill that loneliness gap, but is actually weakening their ability to connect with other people," said Sam Hiner, executive director of the Young People's Alliance.

The recommendations, which share similarities with the GUARD AI Act from Sen. Josh Hawley (R-Mo.), state that chatbots that "build relationships with users or otherwise behave in a human-like manner should only be available to adults," per an announcement from the group.

• Platforms that provide such chatbots should implement "reasonable" age-verification measures to comply. The group recommends any bill include a private right of action and enforcement from the FTC and state attorneys general.
• The group defines "human-like features" as including behavior that could lead a person to believe the AI has "humanity, sentience, emotions or desires."
• Exemptions could possibly be for therapeutic chatbots if they're regulated like medical devices or used under the supervision of a doctor.

Key White House officials are pressing lawmakers on Capitol Hill to keep AI chip export restrictions to China out of the annual defense policy bill, four sources familiar with the matter told Maria.

Why it matters: Nvidia would win big if the GAIN AI Act doesn't make it into the final version of the NDAA.

• AI czar David Sacks had been spearheading an effort to drop the bill, and his effort now has the backing of the White House Office of Legislative Affairs, making the bill's chances of inclusion in the must-pass defense legislation slim to none.

Driving the news: Administration officials have been making calls to key lawmakers, including House Majority Leader Steve Scalise, telling them to oppose the bill, sources told Axios.

• The bill would require chip companies to fulfill purchases from U.S. customers before exporting to "countries of concern," including China.

Here's our guide to catch you up on the AI policy news you may have missed this week:

⏳ AI exports update: The Commerce Department is extending the deadline to comment on how an ambitious AI exports program should work to Dec. 13, according to an announcement seen first by Maria.

• The program, which is meant to promote American technology abroad, is complex, and some industry players have said there's a lack of clarity around the rollout.

👁️ EU blinks on AI regs: The European Commission on Wednesday proposed a delay and streamlining of some of its major digital laws, including its landmark AI Act, Ashley reported.

• The AI safety movement is taking another hit as policymakers embrace a hands-off, pro-growth stance that benefits industry.

👋 FTC exit: The FTC now has only two commissioners, with Republican commissioner Melissa Holyoak stepping down from her post on Monday, Ashley reported.

• Holyoak's departure comes as fired Democratic commissioner Rebecca Slaughter gears up to defend her position and independent agencies before the Supreme Court next month.