In June 2026, the US government placed export controls on a public version of Anthropic's AI coding model, Fable, days after its release. The company revoked access to both that model and its restricted predecessor, Mythos, within hours-sparking debate about whether the move strengthens national security or weakens the country's cyber defenses.
Anthropic built Mythos earlier this year and shared it with a small group of cybersecurity experts, warning it could pose a global cybersecurity threat. The safer public variant, Fable, launched on June 9. That Friday, the federal government intervened, labeling it a national security risk and imposing export controls. Amazon CEO Andy Jassy, whose company both invests in Anthropic and develops rival AI, was the one who alerted officials to the danger. The ban may not survive legal scrutiny-it's unclear that offering API access qualifies as "exporting"-but the fallout is already spreading.
Rethinking reliance on American AI
French politician Bruno Retailleau called the move a "wake-up call" for Europe to build more sovereign AI. Yet any push to turn Paris into a new Silicon Valley runs into a practical obstacle: China. Open-source models from China are highly capable, cheap, and can run on any server with no built-in restrictions. That makes them appealing to companies that don't want their tools suddenly switched off by a White House decision-and equally attractive to the very cybercriminals Anthropic aimed to thwart.
It's a trade-off that could escalate quickly. Shares of Chinese startup Zhipu surged on the news. If US and European firms begin adopting Chinese models en masse, the government's next drastic measure might be to declare those companies a national security threat. The pattern isn't hard to imagine.
A less secure cybersecurity posture
Removing access to Anthropic's models may leave the country more exposed, not less. Leading cybersecurity experts said in an open letter to the government that their teams were using Fable and Mythos to prepare defenses. They argued the models are no more dangerous than other widely available tools. Applying a nonproliferation framework-the kind used for nuclear materials-to software creates a new set of risks, they warned.
Lawmakers scramble to respond
Congress is still catching up. After an earlier feud between Anthropic and the Pentagon over military use of AI, legislators introduced bills defining limits for autonomous systems. Now, each White House intervention amplifies the demand for formal rules. Most Americans want federal AI regulation, polling shows, but lawmakers remain divided on even basic questions like how to govern children's chatbot use or whether the government should vet model safety before release. For professionals focused on AI for Government, the pressure to create coherent policy grows with every executive action.
Why this matters for government
The administration's approach to AI has swung from deregulation under President Trump to labeling America's most valuable AI startup a national security risk twice in one season. That volatility means agency leaders and congressional staffers cannot afford to treat AI as a distant tech sector issue. Understanding the implications of export controls, open-source competition, and cybersecurity preparedness now falls squarely within the remit of government operations. For those shaping policy, structured resources like an AI Learning Path for Policy Makers can help translate technical threats into actionable governance before the next crisis hits.
Your membership also unlocks:
