Most Americans have yet to work out quite what they think about artificial intelligence, but already the AI game is under way: the buying and placing of political pawns in readiness for America’s first comprehensive federal AI law. Voters have yet to elect their senators and representatives, but the power brokers – political donors – are spending, primary by primary, to help ensure their chosen candidates survive to write that all-important legislation.
Two rival networks with differing views about what this law should entail have raised more than $200 million for this year’s midterm elections. Through the end of June, federal filings show, they had put roughly$44 million behind some 40 House and Senate candidates, and, of the candidates they have backed whose primaries have been decided, all but a couple have won.
For anyone who remembers the crypto-currency campaigns of 2024, it’s a familiar playbook. Then the cryptocurrency industry built a super PAC network called Fairshake, funded chiefly by Coinbase, Ripple and the venture firm Andreessen Horowitz, which raised roughly $260 million over the election cycle. A super PAC may spend unlimited sums for or against candidates so long as it does not coordinate with their campaigns.
Fairshake concentrated enormous sums on a small number of races. It put more than $10 million into ending Katie Porter’s Senate hopes in a California primary. The Fairshake network spent about $40 million helping Bernie Moreno defeat Sherrod Brown, the Senate Banking Committee chairman, in Ohio.
Every ambitious legislator in America now knows what it costs to put your name on an AI safety bill
The strangest part was what the advertisements said, which was everything except the word “crypto.” Axios reviewed 59 spots from Fairshake and its affiliates and found ads about insulin prices, border security and China, without a single mention of digital assets. Matthew Erickson, a political messaging consultant, told the outlet that the omission was “usually the sign of a group that’s operating with some sophistication.”
The advertisements praised or attacked candidates on other grounds entirely; their purpose was to seat lawmakers sympathetic to crypto and remove those seen as skeptical. Voters never saw the true argument, but every politician understood what was really going on. Within a year of the election, the crypto industry got its first federal law, governing stablecoins – digital tokens pegged to the dollar. A bill deciding who regulates the rest of the market has since cleared committee in the Senate, where it awaits a floor vote.
The crypto boom showed how quickly election spending could change the personnel and mood in Congress. And it’s some of those same crypto operatives who have moved to AI and into two opposing factions.
In one camp is Leading the Future, the Fairshake-schooled network which launched last year with money from Andreessen Horowitz, OpenAI’s president Greg Brockman, Palantir co-founder Joe Lonsdale, investor Ron Conway and the AI firm Perplexity. It says it raised $125 million in the second half of last year alone, has spent more than $24 million on primaries so far – and that of the 28 candidates it has backed, 25 have won. Leading the Future, the largest AI political network, shares Fairshake’s two-party structure and is co-run by Josh Vlasto, a veteran of the crypto operation.
In the other camp is Public First Action, a nonprofit led by the former congressman Brad Carson, which says it raised about $80 million in June. Anthropic gave it $20 million, though the company says that money is restricted to promoting public education on AI policy rather than electioneering. The group does not disclose its other donors, but Carson says they include employees of OpenAI, Google, DeepMind and X.
The issue that divides Leading the Future and Public First Action is AI safety. For Public First Action, the great danger is AI escaping human control. For Leading the Future, the danger is not moving fast enough, and the bogeyman is China’s AI seizing the century. To understand how they both work to promote preferred candidates and undermine opponents, just look at the case of a young Manhattan assemblyman named Alex Bores.
Last year, New York state passed the RAISE Act, a state law requiring the largest AI developers to publish safety plans. Its Assembly sponsor, Bores, then ran for the congressional seat of retiring Representative Jerry Nadler, and the AI networks’ money followed him into the race.
By one ad tracker’s count, the race became the second most expensive House primary on record. Roughly $27 million in outside spending came from the rival AI factions: committees aligned with the industry, including Leading the Future, spent more than $8 million attacking Bores; committees aligned with AI safety causes, including Public First Action, spent roughly $19 million defending him. He lost anyway.
Bores claimed in a statement after his defeat that the “AI oligarchs” had “set out to make people afraid to stand up to them.” Certainly, every ambitious legislator in America now knows what it costs to put your name on an AI safety bill.
Micah Lasher, who won the congressional primary, said: “I have some news for the two big AI companies who’ve taken such an unusual interest in who won this congressional seat,” before informing both that he would not be taking their cues. Lasher had identified the two poles of the fight, though the money itself traveled through executives, donors, nonprofits and super PACs rather than the companies as such. Interestingly, as in the case of crypto, the attack ads barely mentioned artificial intelligence.
Bores is the only Public First Action-backed candidate to lose. The network has backed ten other winners, supported Scott Wiener, the state senator behind California’s AI transparency law who topped his San Francisco primary in June, and says it will involve itself in 50 to 60 races by November – all in the service of keeping state safety laws alive and subjecting the most powerful models to binding safety frameworks, transparency and incident reporting. “They have a lot of benefits,” Carson told CNBC about the risks of AI. “They have a lot of dangers. And you can’t just release them into the wild with no government concern.”
Each of the rival factions summons a different version of the coming AI apocalypse to raise funds for their cause. The argument that AI could end humanity is not foolish, and the people making it cannot be dismissed as cranks. In 2023, hundreds of researchers signed a statement calling the risk of extinction from AI a global priority, among them Geoffrey Hinton and Yoshua Bengio, two of the most decorated scientists in the field.
They may be right. The possibility warrants research, testing and emergency planning. But a danger can be treated as significant without becoming the premise of how to govern AI. The extinction scenario is politically irresistible precisely because its downside is unlimited and its probability unknowable, so no evidence can ever discharge it. Once catastrophe becomes the premise, questions about cost sound like greed and requests for evidence come off as denial. A premise no one can refute is a priesthood, not a policy.
Americans are not indifferent to AI. Polling this spring found 91 percent of voters want it regulated in some form, and nearly four in five believe it will increase unemployment. Voters also oppose the industry’s flagship demand, federal preemption of state AI laws, by roughly three to one.
A strategist inside the AI industry’s political operation tells me that they have yet to see a race this year won or lost over artificial intelligence, and that lawmakers mostly want answers for constituents about their jobs, their children and their privacy. The political network supplies them and the answers, you will not be surprised to learn, never turn against the technology. That is why this fight runs through low-turnout primaries. Voters want safeguards but have not settled on what they should require or who should write them. The donor networks are spending to settle both questions first.
One of those questions is already being answered. Two years ago, Colorado passed the first comprehensive state AI law in America: a requirement that companies using algorithms to decide who gets hired, housed or loaned money take reasonable care to stop discrimination. In April, Elon Musk’s xAI sued to block it and the Justice Department intervened on the company’s side, the first time the federal government has moved to invalidate a statewide AI law.
Voters want safeguards but have not settled on what they should require — or who should write them
Within days the state agreed to suspend enforcement and in May, Colorado repealed the law, replacing it with a far narrower disclosure measure that has yet to take effect.
The development side has the better argument. America should keep building. Leading the Future’s case rests on the threat of China’s AI advancements, and that part of its argument is true. I spent years in China reporting on the total surveillance state Beijing has built – the cameras, the checkpoints, the algorithms that decide who gets detained. Beijing’s stated goal is to lead the world in AI by 2030. In AI technologies that reshape military and intelligence power, the US cannot easily recover lost ground. Treating an unfalsifiable doomsday prophecy and a hostile, authoritarian police state as equal dangers is a false equivalence.
Both Public First Action and Leading the Future are moving into the late-summer primaries, spreading across the map as they go. The AI safety side says it will be in 50 to 60 races by November. The pro-builder side is grading candidates on scorecards and is no longer confining itself to Congress: it has begun spending in state legislative races, where the next RAISE Acts would be written, and in a governor’s race in Florida.
In January, a new Congress will convene. Many of its members will arrive having learned which AI positions attract money and which attract attack campaigns. It will take up America’s first federal AI rules and decide whether Washington can override individual states. Candidates will know what they should say before the federal debate has begun. The world only has to end once, but that end can be sold every two years.
Comments