US curbs Anthropic AI access, raising global concerns

Silicon Valley executives and global policymakers are scratching their heads at US President Donald Trump's latest crackdown on Anthropic, the artificial intelligence (AI) company behind the powerful Claude models.

Just days after the release of its most powerful upgrades — Claude Fable 5 and the even more powerful Mythos 5 — the Trump administration imposed strict export controls on Anthropic's tools on June 12.

Washington cited national security risks from so-called jailbreaking — clever prompts that bypass AI safety rules — which the San Francisco-based tech giant said were minor, overblown and also present in rival AI platforms.

Yet with the US Commerce Department effectively banning foreign nationals worldwide from using the models, including Anthropic's own staff, the firm had little choice but to suspend global access entirely.

Washington's move came days after Anthropic filed plans for a public listing, likely this fall, hoping to raise tens of billions from investors.

Experts, allies stunned by US 'kill switch'

The export ban drew sharp criticism from the technology sector.

In an open letter published on Sunday, a group of more than 170 tech executives warned that the curbs "risked America's AI leadership" by depriving cybersecurity defenders of their strongest tools, while China's capabilities advance rapidly.

Anthropic's Mythos, Fable blocked after US bans foreign use

To view this video please enable JavaScript, and consider upgrading to a web browser that supports HTML5 video

At this week's G7 summit in Evian, France, Indian Prime Minister Narendra Modi called for "broad and inclusive" access to US AI models, while the UK requested a carve-out from the ban, which was rejected.

Some European lawmakers, meanwhile, described Washington's ability to shut off access as a "kill switch" that reinforced the need for the European Union to secure its own AI sovereignty.

"We won't buy any models made by these companies if overnight, you can just flip the switch," French President Emmanuel Macron warned on Wednesday.

Ad hoc moves expose US regulatory gaps

Risto Uuk, head of European policy and research at the Future of Life Institute, described the US move as "hasty and uninformed" and called on Washington to set clearer and stronger AI regulations similar to those being rolled out by the EU in August.

AI safety "cannot depend on a single firm's goodwill on a given week," Uuk told DW.

Washington has typically deployed lawfare tactics — legal tools like export controls — against foreign rivals such as China and Russia.

Targeting an American company like Anthropic, however, sets a dangerous new precedent, say tech experts and policymakers, as it risks undermining investor confidence in the AI boom, stifling innovation and weakening the US's overall technological edge

Clemens Fuest, the president of Germany's ifo Institute, warned in a research note published Friday that the export ban highlights Europe's "vulnerability" in AI capabilities and called for an expansion of data centers, chip factories and energy infrastructure within the bloc.

Despite being a major AI user, Fuest said Europe controls less than 5% of global AI infrastructure, versus the US with 75% and China with 15%.

Does AI pose a security threat?

To view this video please enable JavaScript, and consider upgrading to a web browser that supports HTML5 video

Anthropic made enemies in the Pentagon

Earlier this year, Anthropic clashed with the US Department of Defense after refusing to lift longstanding curbs on its AI models for the mass surveillance of US citizens or fully autonomous lethal weapons systems.

The Pentagon designated Anthropic a "supply chain risk" and threatened to cancel contracts worth hundreds of millions of dollars, prompting the Claude-maker to launch legal action.

Anthropic has built a reputation for being among the most cautious players in the frontier AI race, adopting a safety-first approach where the models are trained to self-critique against a set of written principles, known as Constitutional AI.

Is Anthropic being over-cautious?

AI researcher Pedro Domingos argues that it is Anthropic, not the Trump administration, that is guilty of dangerous overreach — positioning itself as a self-appointed moral authority.

"They really do honestly believe that AI is a mortal danger," Domingos, a computer science professor at the University of Washington, told DW. "And they also believe that they should be running the world."

"A lot of their problems with the government are that they seem to arrogate [usurp] to themselves the functions of a government," he added.

But will these controversies harm Anthropic's plans for a public listing in the fall?

Anthropic's IPO plans face headwinds

Bankers believe the company could raise $30-$60 billion (€26.2-€52.3 billion), making it one of the largest IPOs of all time. The firm was recently valued at almost a trillion dollars and is now generating revenue estimated at nearly $47 billion a year.

Yet being targeted by the White House, facing export bans on its flagship models and losing out on major government contracts are likely to weigh on investor sentiment.

Wall Street is hoping that Anthropic — along with a potential listing for ChatGPT owner OpenAI — can replicate the extraordinary enthusiasm that greeted Elon Musk's SpaceX IPO earlier this month, which raised $75 billion.

According to IPO expert Jay Ritter, the market believes that the foreign-access curbs on Anthropic's top-tier AI models will soon be lifted, meaning that the standoff won't deter investors.

Citing the Kalshi prediction platform, the University of Florida professor said there was an 85% chance that Anthropic would announce an IPO before November 1, "a percentage that has not changed by very much in the last week."

"There is still enormous enthusiasm in both public and private markets for AI companies," Ritter told DW.

In a pointed rebuke of the US crackdown, Anthropic CEO Dario Amodei told leaders this week at the G7 summit to prioritize international cooperation on AI regulation over unilateral action, urging them to "resist the temptation to splinter."

Edited by: Tim Rooks