WASHINGTON – The U.S. Department of Commerce has issued an unprecedented directive forcing Anthropic PBC to disable access to its most advanced artificial intelligence models for all foreign nationals, marking a significant escalation in the federal government’s control over frontier AI software.
The order requires the immediate suspension of access to the Fable 5 and Mythos 5 models for any foreign national, regardless of whether they are located within the United States or abroad. In response, the developer has shut off access to both systems for all customers to ensure full compliance with the national security mandate, a step that effectively treats large-scale AI models as controlled dual‑use technology similar to advanced chips and cryptographic tools.
This action represents a fundamental shift in U.S. trade and security policy. While previous administrations have utilized export controls to limit the movement of hardware, such as high-end semiconductors and supercomputers, the current directive targets the software weights and access layers of the AI models themselves under the federal export-control regime administered through the Export Administration Regulations. This transition from hardware-centric restrictions to software-centric bans raises immediate commercial and constitutional questions regarding the governance of American intellectual property and the extent of executive authority over cloud-delivered AI services.
Security Vulnerabilities and Corporate Friction
The Commerce Department’s intervention follows the discovery of a “jailbreak” vulnerability in Fable 5, a model designed with specific guardrails to prevent it from executing cybersecurity tasks and assisting with offensive digital operations. Researchers at Amazon.com Inc. identified these vulnerabilities, and Amazon Chief Executive Andy Jassy was involved in direct communications with senior U.S. officials regarding the risks that the model could be repurposed for high-end hacking, malware design, and penetration testing.
The directive has intensified a volatile relationship between the AI developer and the executive branch. Earlier this year, the Pentagon clashed with the company over the application of its technology in military and surveillance operations, according to people familiar with the discussions. Following that dispute, the administration designated the company as a U.S. supply-chain risk, leading to a mandate for federal agencies to phase out its products and limiting its eligibility for future defense and civilian technology contracts.
“We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people,” Anthropic said in its website post. “If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers.”
The administration has countered these claims through David Sacks, co-chair of the President’s Council of Advisers on Science and Technology, who stated that the company refused to remediate the safety issues in the Fable model to the level requested by federal reviewers. Sacks indicated that the export control would be lifted only after the company demonstrates, through a formal technical review process, that the safety issue is resolved and that future updates will remain compliant with the national-security directive.
Market Implications and the Race to IPO
The restriction occurs as the primary players in the AI sector compete to prove the commercial viability of their technology to investors and regulators simultaneously. Both Anthropic and OpenAI are pursuing initial public offerings as soon as 2026, positioning themselves as systemically important infrastructure providers for enterprises, governments, and cloud platforms in the wake of high‑profile listings by other technology and space companies.
The sudden imposition of access controls creates a precarious environment for other major developers, including Alphabet Inc.’s Google and Meta Platforms Inc., which are also experimenting with highly capable foundation models that could fall under similar scrutiny. The directive appears to cut against a recent executive order signed by President Trump that signaled the administration would initially avoid a blanket, mandatory licensing regime for model reviews, instead favoring voluntary disclosures and targeted enforcement.
The economic risks of this policy are summarized below:
| Risk Factor | Business Impact |
|---|---|
| Market Precedent | Creates a template for mandatory government review and potential pre-clearance of all frontier models before release. |
| Revenue Loss | Immediate cessation of foreign subscription and enterprise revenue for affected models, with uncertain timelines for restoration. |
| Competitive Lead | Risk of driving international users toward non-U.S. AI alternatives to ensure continuity and regulatory predictability. |
| Valuation | Potential volatility in IPO pricing and capital-raising plans due to heightened regulatory and political risk. |
For investors, the move underscores that access to advanced AI models is no longer governed solely by technical performance and market demand, but also by a fast-evolving layer of security classifications, export licenses, and compliance audits that can abruptly reshape growth forecasts.
Global Sovereignty and Trade Alliances
The move has triggered an immediate reaction from the European Union, which is currently assessing the statement from Anthropic PBC and its implications for transatlantic data and technology flows. The European Commission stated that these developments highlight the necessity for Europe to establish its own technological sovereignty to avoid dependence on U.S. software that can be revoked or curtailed by executive order with little advance notice.
Aidan Gomez, co-founder of Cohere Inc., described the directive as a “massive wake-up call” for those who believed the U.S. government would not exert such leverage over AI developers. Gomez noted that the action is not “partnerly” and could damage technological alliances and trust frameworks painstakingly built over the last 80 years through trade agreements and shared research programs.
The conflict underscores a tension between the push for global adoption of American AI-supported by leaders like Nvidia CEO Jensen Huang and OpenAI CEO Sam Altman-and a national security framework that increasingly treats high-capability AI as a strategic asset to be guarded under the same logic that governs sensitive encryption, satellite systems, and cyber tools. For allied governments, the directive raises questions about whether access to U.S.-built frontier models will in practice be governed more by security classifications than by commercial contracts.
The U.S. government maintains that the restriction will remain in place until the safety vulnerabilities in the Fable model are remediated to the satisfaction of the Commerce Department and its interagency security partners. Officials have signaled that this order should be read as an early test case for how Washington intends to wield its export-control authority over AI systems, rather than as a one-off dispute with a single company.
Worth a look
