The White House is caught in an Anthropic bind over Mythos

Just months after calling AI safety regulations "foolish rules" and rolling back oversight, the White House is slamming on the brakes.

The reason? Anthropic’s new Mythos model. Its advanced cyber capabilities are powerful enough to spook Washington, turning an ongoing feud into a sudden regulatory pivot. On May 4, The New York Times reported that the Trump administration is discussing an executive order to create a joint industry-government working group to examine pre-release vetting for new frontier AI models.

Throttling compute and threading needles

The most immediate friction isn't just future rules—it's raw compute right now.

Anthropic planned to expand commercial access to Mythos from about 50 firms to nearly 120. U.S. officials pushed back, arguing that doubling private access could strain compute resources needed for government use. Essentially, Washington wants first dibs.

Meanwhile, the administration is preparing a national security AI memo aimed at pushing multi-vendor adoption across agencies. According to Axios, the action would allow federal agencies to sidestep previous "supply chain risk" designations, letting them tap Anthropic's tech despite ongoing legal friction.

Yet internal division remains sharp. While White House officials briefed executives from Anthropic, Google, and OpenAI last week on oversight plans, Sec. of War Pete Hegseth said Thursday that Anthropic is "run by an ideological lunatic."

A sharp reversal from hands-off AI

This marks a stark shift for an administration that previously championed total deregulation.

After taking office, President Trump scrapped Biden-era reporting requirements for high-risk models, arguing the tech was vital to beating China. Last July, he called AI a "beautiful baby" that shouldn't be stopped by politics or stupid rules. But sources told The New York Times that officials now fear the political fallout if a devastating AI-driven cyberattack were to happen.

Officials are now looking at models similar to the UK system, where state bodies vet AI systems against safety standards. Former AI czar David Sacks noted that GPT 5.5 has already matched Mythos's cyber capabilities, predicting all frontier models will hit that mark within six months.

Why it matters

Washington's change of heart proves that absolute laissez-faire AI policy vanishes the moment national security risks become tangible.

For businesses waiting on Mythos, the immediate hit is access. The government rationing private compute to safeguard its own pipeline means enterprise rollouts will face bureaucratic bottlenecks. Developers should expect pre-release security reviews to become standard across top labs.

Strategically, the White House's approach looks messy. Trying to patch relations with Anthropic to use Mythos while simultaneously proposing new vetting hoops signals internal incoherence. Furthermore, as Sacks pointed out, these cyber capabilities will be widespread across all frontier models in six months. Attempting to manage model safety by hoarding compute and holding informal meetings is an analog solution to an exponential problem.