San Francisco-based Anthropic watched its newest models vanish from customer use last week. The Trump administration ordered the company to block foreign nationals from accessing Fable 5 and the more powerful Mythos 5. National security concerns drove the decision. Compliance required Anthropic to pull both systems offline entirely.
Employees reacted with confusion. Internal chats captured the uncertainty. “What are you telling your clients?” one asked. Another wondered, “Does anyone know what to believe?” The New York Times reported these exchanges on June 17, 2026. Executives including CEO Dario Amodei met with administration officials that same day. No resolution emerged. The company issued a statement pledging continued cooperation.
But the episode runs deeper. It marks the latest clash in a year of tension. Earlier, the Pentagon labeled Anthropic a supply-chain risk after the firm resisted certain military uses of its technology. Lawsuits followed. A federal judge issued a preliminary injunction blocking parts of that action. The pattern suggests more than isolated security worries. It points to a government testing boundaries on frontier AI control.
The Trigger: A Jailbreak and Export Controls
The immediate cause traces to a vulnerability. Amazon researchers identified a way to bypass some of Fable’s cybersecurity guardrails. Commerce Secretary Howard Lutnick sent a letter to Amodei. The directive applied export controls to the models for any location outside the U.S. and all foreign persons inside it. Deemed-export rules left Anthropic no choice but to suspend access broadly.
The Fortune article from June 16 framed this as an ad hoc licensing regime. Officials repurposed existing authorities. Jonathan Iwry of the Wharton Accountable AI Lab called it a “backdoor licensing regime.” Dean Ball, who once advised the administration on AI policy, criticized the move sharply. The government now effectively decides which models can reach which users. Yet it avoids formal acknowledgment of that power.
Anthropic had positioned itself as safety-first. Amodei long advocated for regulation. The firm voluntarily limited access to its most advanced systems. It shared models with government evaluators ahead of release. Those steps did not shield it. The administration cited unspecified national-security risks. Cybersecurity experts questioned the proportionality. A group of them urged officials to ease restrictions, arguing the action could benefit adversaries more than it protects American interests. The Associated Press covered that appeal on June 17.
So the company that asked for rules now faces their blunt application. Executives scrambled to Washington. Meetings occurred on June 15 and continued. Politico reported the first in-person sit-downs aimed at a truce. One White House official noted resolution might take more than a few days. But the door stayed open. By the G7 summit in Évian-les-Bains this week, talks remained active. Trump told reporters they were “going fine.”
The timing amplified global ripples. AI leaders including Amodei, OpenAI’s Sam Altman and Google DeepMind’s Demis Hassabis joined world leaders at the summit. European officials expressed alarm. French President Emmanuel Macron pushed for a “trusted partners” framework. Allies worry about sudden loss of access to American models. Ursula von der Leyen called for deeper transatlantic cooperation on AI safety standards. The episode underscores how U.S. decisions now shape allied technology strategies.
Yet domestic confusion persists. NBC News reconstructed the scramble inside the administration. A passing reference during a call between Amazon CEO Andy Jassy and officials set events in motion. Disagreements over guardrails and a specific workaround led to the broad order. The NBC report from June 16 detailed how quickly policy hardened.
Anthropic’s blog post confirmed the suspension. It emphasized the firm did not view the flagged issue as warranting such steps. The company remains in dialogue. No public details explain the exact vulnerability or why foreign-employee access posed an insurmountable problem. That opacity fuels criticism. Observers see an administration wielding export controls as a de facto veto on model deployment. The Axios scoop on June 12 first broke the blocking order.
Longer-term questions loom. Frontier AI development demands enormous capital and data. Only a handful of U.S. firms compete at the highest level. Government intervention at this scale risks chilling investment. It also raises consistency issues. Other labs release powerful systems without similar intervention. The selective pressure on Anthropic, given its prior disputes, invites accusations of targeting.
At the same time, security cannot be ignored. Advanced models can generate code, manipulate systems or assist in planning. Jailbreaks erode safeguards. The administration argues it must act when risks surface. But the mechanism—sweeping deemed-export application—forces companies into binary choices. Partial access becomes impractical. Models go dark for everyone.
Amodei used the G7 stage to urge unity. He warned against splintering among democracies. Altman backed the call. Both stressed that private firms cannot shoulder regulation alone. Democratic governments should set standards and share defensive tools. Their rare alignment highlighted shared stakes. Yet the immediate friction with Washington overshadowed the message.
Negotiations continue. Trump’s comments suggest optimism. Commerce officials hint at possible adjustments. For Anthropic, restoring access matters commercially. Fable and Mythos represent its latest leap. Customers wait. Enterprise contracts hang in balance. The firm’s valuation, near $1 trillion according to some estimates, reflects confidence in its trajectory. That confidence now collides with policy reality.
The episode reveals broader tensions. Industry leaders once lobbied for oversight. They got it. But the form—opaque, rapid, tied to national-security levers—differs from the thoughtful frameworks many envisioned. Congress has not passed comprehensive AI legislation. Executive actions fill the void. The result feels improvisational.
Experts debate long-term effects. Some see necessary caution against proliferation risks. Others warn of self-inflicted wounds. Allies may accelerate their own models or turn elsewhere. Domestic innovation could slow if firms fear sudden shutdowns. And the public receives mixed signals. One day officials tout American AI dominance. The next, they sideline a flagship product.
Anthropic insists it will keep working with the administration. Amodei and his team emphasize safety commitment. They have paused releases before to address risks. This case differs because the government forced the pause. The distinction matters. Voluntary restraint builds trust. Mandated withdrawal breeds uncertainty.
As talks proceed, the Fable and Mythos models remain unavailable. Employees field client questions without clear answers. Policymakers weigh security against economic and diplomatic costs. The outcome will shape how the U.S. governs its AI edge. For now, the approach stays experimental. One company bears the immediate burden. The industry watches closely. Allies voice concerns. And the technology marches forward, constrained but not contained.
Recent reporting adds texture. The Politico story from June 15 described the meeting dynamics. White House officials left room for quick fixes but tempered expectations. CNBC noted the irony of a regulation-friendly company facing aggressive enforcement. NPR highlighted regulatory confusion. These accounts, published in the days surrounding the New York Times piece, show a story still unfolding. No single explanation satisfies all parties. The gap between stated security needs and practical impact remains wide.
WebProNews is an iEntry Publication