Back to News
RIA Novosti

When Washington pulls the plug: The new front in AI control

Last week, Anthropic—the AI company behind the Claude family of models—temporarily restricted access to its two most advanced systems, Claude Opus 4 and Claude Opus 4 Turbo. On its face, it looked...

Last week, Anthropic—the AI company behind the Claude family of models—temporarily restricted access to its two most advanced systems, Claude Opus 4 and Claude Opus 4 Turbo. On its face, it looked like routine product management. But behind the scenes, something far more significant was unfolding.

According to sources familiar with the matter, the shutdown wasn't Anthropic's idea. The U.S. government requested the move, citing national security concerns. Officials were worried that a newly discovered jailbreak—a method to bypass safety guardrails—could allow the models to be used for finding software vulnerabilities. While Anthropic has stated that the exploit only revealed relatively simple, already-known flaws, and that other leading models are similarly capable, the administration decided to act.

What makes this episode remarkable is that it marks the first time Washington has moved to restrict access to AI *capabilities* themselves, rather than just the hardware that powers them. For years, export controls focused on advanced chips. Now, the target is intelligence.

Anthropic’s relationship with the federal government has been tense. The company’s policy prohibits using its Claude models for mass surveillance or lethal autonomous weapons—a stance that has drawn criticism from the Pentagon and, notably, from former President Donald Trump, who labeled the company “radical left.” In February, several federal agencies were reportedly advised to phase out Anthropic’s tools. The company has not confirmed the scope of that directive.

The irony is not lost on industry observers. AI developers spent years warning governments about the growing power of their systems. Now those same governments are listening—and clamping down.

It’s too early to say whether this is a one-off incident or the start of a new regulatory era. But if the trend holds, access to the most capable AI models may soon depend less on your budget and more on your passport. Nations that build their own frontier AI systems will gain a strategic edge; those that rely on imports may find themselves locked out of the very tools that define the next decade.

For now, Anthropic says it hopes to restore access within weeks. But the precedent has been set: when a government decides a model is too powerful to share, it can pull the plug.