The White House Is Making Up Its Rules for AI in Real Time

A Directive Without Clear Rules

Nearly a week has passed since the Trump administration sent an export control directive to Anthropic, forcing one of the world's leading AI labs to pull its most advanced models offline. After days of negotiations, the two sides remain at odds over how to bring Claude Mythos and Fable 5 back. Anthropic does not believe it violated any concrete procedures, while the White House contends the company acted recklessly.

The Wild West of AI Regulation

This saga confirms that the US has entered a Wild West era of AI regulation. Few laws govern frontier AI development, but companies can still run afoul of unspoken lines. A former White House technology official, speaking anonymously, said: "The problem here is that the White House has been in this extreme anti-regulatory posture, and they’re now faced with the real AI capabilities… There should have been preparation and policies, but instead it's just this slap-dash approach."

The Trump administration has repeatedly blocked efforts to impose AI guardrails, arguing they could hamper innovation. President Trump has signed executive orders reversing Biden-era AI frameworks and created a task force to challenge state laws.

Opaque Dispute and Its Consequences

The dispute remains opaque – the US government has never clearly stated what Anthropic did wrong, with only a social media post from White House tech advisor David Sacks outlining the situation. Ironically, the White House's actions may have hampered the innovation it aims to protect. The administration demanded Anthropic ban all foreign nationals from accessing Mythos and Fable 5, locking out many of its own employees and all customers, including Apple, Meta, and much of the Fortune 500.

Legitimate Concerns?

The White House may have had legitimate worries. US officials grew concerned when they learned Anthropic shared Mythos with SK Telecom, a South Korean telecom giant alleged to have ties to China. Separately, Amazon CEO Andy Jassy raised concerns about potential jailbreaking of Claude Fable 5's safeguards. Anthropic says it coordinated with the government on the rollout and revoked access immediately after being alerted.

The Real Problem: Ad-Hoc Decisions

Jailbreaking is an inherent issue in all large language models – they are probabilistic, not deterministic, so no guarantee can be made. If the White House insists on solving jailbreaks before allowing Fable 5 back, it could take a long time.

Impact on the Industry

Other AI labs like OpenAI, Google, and Meta are watching closely. Many AI leaders now believe they must give the White House early access to their latest models and proactively share information about upcoming launches. President Trump's executive order last month created a "voluntary" system for early testing, but after this incident, it effectively becomes a mandatory licensing regime. As the former official put it: "The Trump administration, frankly, should not have said that this was a voluntary regime. It seems very clear that what they are now doing is a licensing regime."