US AI export controls: Anthropic forced to shut down Mythos 5 everywhere

The United States has imposed a hard stop on the expansion of the most advanced artificial intelligences with an unprecedented intervention that led to the blocking of Anthropic’s Mythos 5 and Fable 5 models for all foreign users. This measure, without the possibility of differentiating access according to users’ nationality in real time, pushed the company to shut down both systems globally, marking an important chapter in the debate on US controls on AI exports.

Key points

• The US Department of Commerce has issued a directive that bans access to Anthropic’s Mythos 5 and Fable 5 models for any foreign user.
• Anthropic has globally shut down the models in order not to violate the ban, since it cannot segment users with national-level precision.
• The government intervention came after Amazon reported potential risks of bypassing protections in Mythos AI systems.
• Senior US officials, including Commerce Secretary Howard Lutnick and Treasury Secretary Scott Bessent, attempted mediation with CEO Dario Amodei before issuing the blocking order.
• The measure reflects the application of stricter regulations on frontier AI after failed attempts at voluntary cooperation with companies.

US Government Blocks Anthropic’s Mythos 5 and Fable 5 Models

At 5:21 p.m. on June 12, 2026, Anthropic received a formal order from US Commerce Secretary Howard Lutnick: immediately block access to its Mythos 5 and Fable 5 artificial intelligence models for any foreign citizen, regardless of where they are in the world, including the company’s own foreign employees. The technical impossibility of segmenting users by nationality in real time led Anthropic to shut down the models globally, while keeping less advanced models such as Opus 4.8 active.

The Mythos model stands out for its autonomous ability to identify and exploit cyber vulnerabilities, including zero-days, representing an extremely serious risk for global cybersecurity if it were to fall into the wrong hands. The potential severity of this capability motivates the authorities’ harsh response, which has adopted an export-control approach normally reserved for military or national security technologies.

Government-Company Engagement Before Enforcement

Before the order, several high-level consultations took place. Two days after the public launch of Fable, Amazon CEO Andy Jassy raised concerns with the US government about possible gaps in the safety barriers of the Mythos model. These concerns were shared with the National Security Agency, reinforcing the belief that the models posed concrete risks.

On Friday morning, a high-level meeting was convened with officials such as Commerce Secretary Howard Lutnick, Treasury Secretary Scott Bessent, and White House Cybersecurity Director Sean Cairncross. Despite three calls with Anthropic CEO Dario Amodei, in which he downplayed the issues by arguing that it was a limited violation and not a universal jailbreak, the officials remained unconvinced and requested a voluntary suspension of the model to close the gap. Not obtaining sufficient guarantees, they imposed the block through export control.

According to government sources, this measure was considered a last resort after many hours of attempted collaboration, while according to Anthropic the time granted was extremely short and lacked transparency, fueling controversy over the method.

Anthropic’s Advocacy for AI Regulation and Paradoxical Outcome

CEO Dario Amodei had long been an advocate of a strict regulatory system for artificial intelligence, comparing the management of frontier AI to that of aircraft, with mandatory tests and certifications by third-party authorities. In his essay “Policy on the AI Exponential” he himself identified frontier models as possible “public safety threats” to be blocked or revoked if they did not comply with safety standards, also proposing strict export controls in a strategic and geopolitical key.

This very vision made Anthropic’s reaction to the government block paradoxical and bitter: a company that had publicly highlighted the risks of its own creations found itself surprised and in difficulty when the state actually applied the measures it had itself suggested. In addition, the company admits that safeguards could still be bypassed, anticipating that a universal jailbreak is inevitable in the medium term, an element that likely contributed to increasing the authorities’ concern.

Regulatory and Legal Implications for Frontier AI Models

The block represents a clear signal from the United States of a tightening in the regulation of frontier artificial intelligences, especially when attempts at voluntary cooperation fail. On June 2, days before the measure against Anthropic, the White House issued an Executive Order that invites the most advanced labs to subject their models to confidential safety benchmarks, to which the government must have access up to 30 days before commercial launch.

In parallel, the legal dispute between Anthropic and the government has been ongoing for months. For example, in March a judge blocked the Pentagon from labeling the company as a “supply chain risk,” recognizing free speech issues raised by the company. Now export control could open a new legal phase that will define the limits of state power in AI matters. These developments will be crucial in establishing the balance between technological innovation, national security, and regulatory policy in the United States and beyond.

This episode shows how AI regulation in the US is rapidly evolving toward more rigorous and centralized approaches, especially for the most advanced technologies that can have strategic and global security impacts. For investors and researchers, it is a clear indication that AI governance will require not only technical attention, but also commitment to compliance policies and transparency with the authorities.

Content created with the assistance of artificial intelligence and with human editorial review.