The shutdown affects one of Anthropic’s highest-value product lines. The report says input tokens for Mythos cost twice as much as those for the company’s lower-powered Opus 4.8 model, making the lineup an important part of its revenue plans ahead of a potential initial public offering.
The disruption also comes as Anthropic carries a major infrastructure commitment. The company has agreed to pay SpaceX $15 billion annually for data center capacity, with Mythos revenue expected to help support that expense. The report also notes that Google and Amazon, both significant Anthropic investors, have spent the past year managing their relationships with the White House and may have little interest in seeing the shutdown continue.
The reason negotiations have stalled remains uncertain. According to the report, export controls for AI models do not yet follow an established framework, unlike many traditional dual-use technologies that are typically reviewed during manufacturing well before they reach customers. Instead, Anthropic is navigating what the report describes as a rapidly evolving regulatory process.
The Verge previously reported that the U.S. Department of Commerce evaluated Fable 5 before its release without raising objections. The report says the agency later acted after Amazon CEO Andy Jassy reportedly highlighted a technique for bypassing some of the model’s guardrails. Katie Moussouris, founder and CEO of Luta Security, reviewed the underlying findings at Anthropic’s request and argued the issue had been overstated. According to the report, researchers found the model would reject requests to review code for security issues while responding to prompts asking it to repair code, potentially exposing vulnerabilities through that process.
Anthropic has also changed its negotiating team. Cofounder Tom Brown has taken over as lead negotiator from CEO Dario Amodei and is working alongside public policy chief Sarah Heck. The report says progress has been limited despite Anthropic’s recent efforts to strengthen its relationship with the Trump administration by emphasizing Mythos’ cybersecurity capabilities.
The situation extends beyond Anthropic because other major AI developers could face similar scrutiny. The report notes that OpenAI’s GPT-5.5 Cyber recently outperformed Mythos 5 on certain benchmarks and that the administration has also asked OpenAI to postpone the broader release of GPT-5.6 while planning to approve customer access individually.
The uncertainty has also prompted discussion outside the United States. According to the report, some countries have begun advocating for non-American AI systems over concerns that access to U.S. models could be withdrawn unexpectedly. Alex Stamos, chief product officer at Corridor, said the current situation benefits overseas competitors. “One of America’s champions is being kneecapped by the US government while we’re in a race with the Chinese. It’s just incredibly stupid.”
The report also highlights what it describes as a shift in how frontier AI companies may view regulatory risk. It argues that developers had largely expected advanced model launches to be governed by internal safety testing and voluntary commitments. The current export control process, however, introduces the possibility that a government review could suspend access to a deployed model, creating new uncertainty for companies preparing future releases and public market debuts.
This analysis is based on reporting from The Verge.
Image courtesy of Yahoo.
This article was generated with AI assistance and reviewed for accuracy and quality.
