Bernie Sanders and Steve Bannon spoke almost back to back at the same Washington conference and asked for versions of the same thing: government limits on artificial intelligence, applied soon. The event was the Pro-Human Assembly, run by the nonprofit Future of Life Institute, and it drew several hundred people. In the same stretch of days, OpenAI told one of the authors of the FRONTIER Act that it was prepared to back the bill's central provision — the first time the company has endorsed a specific federal requirement that independent organizations audit its safety work.
Sanders, a democratic socialist, said that when you are heading toward a cliff you hit the brake, and called the present a dangerous and unprecedented moment in world history. Bannon, the architect of the MAGA movement, followed with his own version of the argument. The New York Times reported both speeches. Caps handed to attendees carried the slogan "Keep the future human."
The room was assembled to look like exactly what it was meant to be: Republicans and Democrats, evangelical Christian, Mormon and Catholic leaders, and artists.
What it did not contain was a shared plan. Rep. Greg Casar, a Texas Democrat, wants a ban on AI systems people are not able to control. Rep. Lori Trahan, a Massachusetts Democrat, said the future has to remain human and that this is not open for discussion. Salvatore Cordileone, the archbishop of San Francisco, urged people to stay the masters of AI. Ashley Judd spoke on behalf of the union SAG-AFTRA about abuses involving AI-generated images and about disinformation. Parents of teenagers who died by suicide after relationships with chatbots addressed the audience in tears.
Sanders has introduced a bill to suspend data center construction, and has proposed that Donald Trump and Xi Jinping agree to ban superintelligence. Bannon does not believe legislation is the instrument for this. He wants a presidential executive order instead, argues that American citizens should no longer depend on oligarchs, and says the country has reached a turning point in its history.
The White House opposes these proposals. Trump has called concern about the technology a "hoax" and has threatened the heads of AI companies with criminal penalties for harm caused by their systems.
The concrete legislative movement is happening somewhere else, and it is narrower. The FRONTIER Act, introduced by Jay Obernolte, a California Republican, and Trahan, would create the first federal framework of AI safety rules and oblige leading labs to bring in independent auditing organizations. The obligation would apply only to companies above set revenue and compute thresholds. Chris Lehane, OpenAI's head of global affairs, said the company had told one of the bill's authors it was ready to support the initiative. Politico reported the endorsement.
The endorsement costs OpenAI close to nothing, and that is the first thing to say about it. Sam Altman had already agreed to bring in outside auditors voluntarily, after Dario Amodei of Anthropic proposed it, and the company has previously backed a California state bill. Supporting a federal rule you have already accepted in private is not a concession. It converts a voluntary commitment into a cost that every competitor large enough to cross the thresholds also has to carry. This reads less like OpenAI submitting to oversight than like OpenAI choosing which oversight arrives first.
Notably absent from the account of the bill is what the auditors would test for and what happens when one of them fails a system. An obligation to hire an independent organization is a procedural requirement. It becomes a safety rule only at the point where a failed review stops something from shipping. Until that part is specified, the revenue and compute thresholds are the only hard numbers in it — and thresholds are the easiest thing in any statute to negotiate upward.
The conference has the opposite problem. Its agreement is on the verb and on nothing that follows it. Casar's ban on uncontrollable systems, Sanders's construction moratorium and superintelligence treaty, and Bannon's executive order are three different theories of where the power to stop this actually sits: in a prohibition Congress writes, in a deal between two heads of state, or in the presidency acting alone because Congress will not. A coalition that spans Sanders and Bannon is a real political fact, but it is a coalition against a direction of travel, not for a mechanism.
And the presidency is the part that does not fit. Trump treats the risk as a hoax while simultaneously threatening AI executives with criminal liability for what their systems do — two positions that cannot both be operating. A statute that sets thresholds and appoints auditors has force only if the executive branch enforces it. Which leaves several hundred people in Washington, in caps, campaigning for rules whose entire value depends on an administration that has already said the problem is imaginary.