The Trump administration has officially announced the removal of stringent export controls on two of Anthropic’s most sophisticated artificial intelligence models, Fable 5 and Mythos 5, signaling a significant shift in the federal government’s approach to regulating high-capacity dual-use technology. The decision, formalized in a letter from Commerce Secretary Howard Lutnick to Anthropic co-founder Tom Brown, concludes a period of intense negotiation between the San Francisco-based AI firm and the Department of Commerce. The move effectively ends a licensing requirement that had previously restricted the distribution of these models to a narrow selection of pre-approved government agencies and specific corporate partners, allowing Anthropic to compete more broadly in the global marketplace.
According to the correspondence viewed by WIRED, the Department of Commerce determined that a license is no longer required for the export, re-export, or in-country transfer of the Mythos or Fable models. This includes "deemed exports," a regulatory category that covers the release of controlled technology to foreign nationals within the United States. The lifting of these barriers follows a comprehensive deal in which Anthropic has committed to heightened transparency and a collaborative framework for safety testing with federal authorities.
The Framework of the Agreement
The resolution of the dispute centers on a series of concessions and commitments made by Anthropic to address the administration’s concerns regarding national security and the potential misuse of advanced AI. Central to the agreement is Anthropic’s pledge to proactively detect and mitigate security risks associated with the models’ outputs. Secretary Lutnick’s letter noted that the company has agreed to work "diligently" with the U.S. government to establish protocols and standards for the release of Mythos, Fable, and all subsequent models developed by the firm.
Sources familiar with the negotiations indicate that the White House and the Commerce Department were particularly concerned with the potential for users to bypass safety filters—a process known as "jailbreaking." Specifically, officials feared that the advanced reasoning capabilities of Mythos 5 could be leveraged by adversarial actors to enhance offensive cybersecurity operations, develop biological or chemical agents, or conduct sophisticated disinformation campaigns. To address this, Anthropic has agreed to implement more robust safeguards designed to prevent the models from providing actionable information on restricted topics, even when faced with complex adversarial prompting.
A Chronology of the Regulatory Standoff
The path to this agreement was marked by months of tension and a notable shift in Anthropic’s corporate diplomacy. The conflict began in late 2025, when the Bureau of Industry and Security (BIS) within the Commerce Department flagged the Mythos model as a potential risk to national security due to its unprecedented performance on coding and logic benchmarks.
Initially, Anthropic, led by CEO Dario Amodei, resisted the imposition of strict export controls. The company argued that the administration’s fears were disproportionate to the actual risks and that the technical nature of large language models (LLMs) made it impossible to guarantee a "zero-jailbreak" environment. This stance led to a stalemate, with federal officials expressing frustration over what they perceived as a lack of cooperation from the company’s leadership.
The timeline of the dispute highlights the escalation and eventual resolution:
- October 2025: The Department of Commerce issues a temporary stay on the export of Mythos 5, citing concerns over its "unbounded" capabilities in cybersecurity domains.
- January 2026: Anthropic files a formal challenge, arguing that the restrictions hamper American competitiveness against foreign state-backed AI initiatives.
- March 2026: Reports surface of a "personality clash" between the White House and Anthropic CEO Dario Amodei. National Cyber Director Sean Cairncross and Secretary Lutnick reportedly found the company’s communication style overly academic and dismissive of security concerns.
- May 2026: Anthropic shifts its strategy. Co-founder Tom Brown takes a primary role in negotiations. The company pivots from arguing the impossibility of perfect safety to promising a "best-in-class" mitigation strategy.
- June 2026: A deal is reached, resulting in the letter from Secretary Lutnick lifting the restrictions just ahead of the third-quarter fiscal reporting period.
The Role of Corporate Diplomacy and Leadership Shifts
A critical factor in the administration’s decision to lift the controls was a change in how Anthropic interfaced with government officials. Internal reports suggest that the Trump administration found Tom Brown to be a more pragmatic interlocutor than Amodei. This shift in leadership during the negotiation phase allowed for a reset in the relationship between the private sector and the regulators.
By moving away from the conceptual argument that jailbreaks are an inherent, unsolvable flaw of LLMs, Anthropic adopted a posture that prioritized the government’s desire for proactive risk management. The company effectively chose to align its messaging with the administration’s "security-first" rhetoric, promising to build "defensive moats" around its most powerful models. This pragmatic approach—providing the administration with the technical assurances it sought—was instrumental in breaking the regulatory logjam.
Technical Implications: Understanding Fable 5 and Mythos 5
The models at the heart of this controversy represents the current frontier of generative AI. While Fable 5 is designed as a highly efficient, versatile model for enterprise use, Mythos 5 is a significantly larger and more capable system, possessing advanced capabilities in multi-step reasoning and autonomous problem-solving.
The primary concern of the National Cyber Director was the model’s ability to discover "zero-day" vulnerabilities in software code. In earlier testing phases, restricted versions of Mythos 5 demonstrated a high proficiency in identifying and exploiting architectural flaws in legacy digital infrastructure. Under the new agreement, Anthropic will integrate "safety-by-design" principles that involve continuous monitoring of user queries and the use of secondary "monitor models" to flag and block suspicious activity in real-time.
Furthermore, the deal includes a provision for the U.S. AI Safety Institute to have early access to future Anthropic models before they are deployed internationally. This "red-teaming" collaboration ensures that the government can evaluate the potential impact of new technologies on national security before they become widely available.
Broader Economic and Geopolitical Impact
The lifting of these export controls has profound implications for the global AI landscape. As the United States competes with China and other nations for AI supremacy, the ability of American firms to export their most advanced products is seen as a key component of national power.
Data from the AI Policy Institute suggests that export restrictions can lead to a "market vacuum" where foreign competitors, often with fewer safety guardrails, step in to provide services to international clients. By allowing Anthropic to export Mythos 5 and Fable 5, the administration is betting that American-led safety standards will become the global norm.
Market analysts anticipate that this move will significantly boost Anthropic’s revenue potential, particularly in the European and Middle Eastern markets, where demand for high-end AI integration is surging. It also sets a precedent for other AI giants, such as OpenAI and Google, who are navigating similar regulatory hurdles. The "Lutnick-Brown" deal may serve as a blueprint for how the private sector and the government can co-manage the risks of dual-use technologies without stifling innovation or market expansion.
Official Responses and Industry Reaction
While the Commerce Department has framed the move as a victory for both security and American industry, reactions from Capitol Hill have been mixed. Pro-tech advocates have praised the decision, noting that over-regulation of AI could lead to a "brain drain" and the migration of talent to more permissive jurisdictions.
"This agreement demonstrates that we do not have to choose between national security and American leadership in technology," said a spokesperson for the Department of Commerce. "By working with Anthropic to establish clear safeguards, we are ensuring that the most powerful tools in the world are developed and distributed responsibly."
Conversely, some security hawks have expressed caution. Critics argue that once a model’s weights are exported or accessible via API in certain jurisdictions, the ability to prevent "jailbreaking" becomes significantly more difficult, regardless of the safeguards in place. They emphasize that the "proactive detection" promised by Anthropic must be backed by rigorous, independent auditing to be effective.
Analysis: The Future of AI Governance
The resolution of the Anthropic dispute marks a pivotal moment in the "America First" AI strategy. It highlights a preference for bilateral agreements between the state and individual corporations over broad, industry-wide legislative mandates. This approach allows the administration to remain flexible, adjusting restrictions based on the specific capabilities of a model and the perceived cooperativeness of the company’s leadership.
However, the long-term viability of this strategy remains to be seen. As AI models become increasingly decentralized and open-source alternatives continue to improve, the effectiveness of export controls on proprietary models may diminish. For now, the deal with Anthropic serves as a temporary truce in the ongoing struggle to balance the immense economic potential of AI with the existential and security risks it may pose.
As Anthropic prepares to roll out Fable 5 and Mythos 5 to a global audience, the tech industry will be watching closely to see if the promised safeguards hold up under the pressure of real-world use. The success or failure of this regulatory experiment will likely dictate the future of AI exports for years to come.
