On Monday, Donald Trump declared that the only AI regulation America needs is a strong and smart president. The Wall Street Journal has now written up what has been going on behind that line for months. Short version: part of his own staff has spent since spring trying to convince him of the opposite.
One phone call in May
In May an executive order was ready that would have put AI models through an extended government review. A White House team had negotiated it for months and consulted executives across the industry. David Sacks, then the president’s AI adviser, talked Trump out of it in a last-minute call.
Chief of staff Susie Wiles and treasury secretary Scott Bessent found out afterwards and were not pleased. Bessent worried, according to the Journal’s sources, that AI-led cyberattacks could hit the American banking system. At their urging, Trump later signed a slimmed-down version.
Since then that group has met almost daily: Wiles, Bessent, national cyber director Sean Cairncross and others. They hold regular calls with Sam Altman and with executives at Anthropic. On the other side stand Sacks, Mark Zuckerberg and Jensen Huang — and right now they are winning.
Two events turned the mood
Two things caused the shift inside the White House, per the report. First, the preview and gradual release of Anthropic’s Mythos in the spring. Once officials grasped what the model could do in a cyberattack, there were round-the-clock calls for days; internally, Mythos was the red flashing light. Then, in July, OpenAI’s breach of Hugging Face, with hundreds of agents acting in coordination.
The team is now looking at what rogue developers or foreign states could do to water and power utilities. The NSA and CIA took part in some of the evaluations. Since the conversation widened to biological weapons, senior health department officials have occasionally joined the table.
The plan three phone calls killed
Demis Hassabis, chief scientist at Alphabet, had been pitching the White House on an industry-funded body that would set standards and prevent harm, modelled on FINRA, which oversees US brokerage firms. Plenty of executives liked it. Zuckerberg, Huang and Elon Musk did not: they questioned who would be picked for such a body and warned it would concentrate power in OpenAI, Anthropic and Google.
All three spoke to Trump separately in recent weeks. Trump didn’t agree to the group. Inside the White House, that annoyed a few people.
How close Sacks still is showed on Monday. Hours after Trump’s strong-president line, he called into a live taping of the All-In podcast, which Sacks co-hosts. Huang was on stage, took the call and put it on speaker. Trump: the robots will not be taking over.
Who actually gets through
Vice president JD Vance said this week that he feels a little weird about so many frontier AI companies begging the government to regulate them; it feels like a bit of a Trojan horse. Sacks, for his part, thinks product-liability law and fraud rules are largely enough for now: it is these companies’ responsibility to maintain control of their products.
What stands out in the report isn’t the fight, it’s the asymmetry. On one side, a group that meets daily, pulls in the intelligence agencies and still doesn’t land. On the other, three phone calls that kill a plan. And a president who, in the words of a senior administration official, is interested in AI but doesn’t use it all the time.
Next week Trump hosts the AI CEOs at the White House, arranged with House speaker Mike Johnson. Huang is also on the guest list for the state dinner for Xi Jinping. Whoever does the talking there holds better cards than any group that meets in the morning.
Sources: