The United States government has lifted its export restrictions on Anthropic's two most powerful artificial intelligence models, Claude Fable 5 and Mythos 5, less than three weeks after ordering their global suspension over national security concerns. The San Francisco-based AI company announced late Tuesday that it would begin restoring access to both models the following day, after the US Department of Commerce formally notified it that export controls had been removed.

The restrictions had been imposed on 12 June following a government order that required Anthropic to cut off all foreign access to the models — including access by the company's own non-US employees. Anthropic said at the time that officials had not provided specific justifications beyond vague national security concerns, though the company understood the government to be worried about a method of bypassing the models' safety restrictions, known as "jailbreaking." In a letter to Anthropic, Commerce Secretary Howard Lutnick confirmed the lifting of controls, stating that the company had agreed to "proactively detect and address security risks" in the models, collaborate with the government on protocols for future releases, and alert authorities to any malicious activity. Lutnick also noted that the department reserves the right to reimpose restrictions should circumstances change or Anthropic fail to meet its commitments. A company source confirmed that Anthropic had implemented a new safeguard specifically targeting the jailbreak technique the government flagged, though the source noted the underlying vulnerabilities had already been identified and patched before the order was issued.

Fable and Mythos, both built on Anthropic's Claude platform — a rival to OpenAI's ChatGPT and Google's Gemini — share the same underlying AI architecture but serve different purposes. Fable 5 is a consumer-facing model capable of complex independent reasoning, while Mythos 5 is designed for businesses and cybersecurity professionals, with some safety guardrails relaxed to allow it to identify and exploit vulnerabilities in computer code. Experts had warned that Mythos in particular could significantly accelerate sophisticated cyberattacks if it fell into the wrong hands.

The episode is part of a broader tightening of US government oversight over frontier AI models, driven by concerns about potential misuse by military or intelligence services in China, Russia, and other countries. President Donald Trump signed an executive order establishing a voluntary framework under which AI developers must offer advanced models to the government for review for up to 30 days before wider release. OpenAI, meanwhile, delayed the full public launch of its latest GPT-5.6 model at the government's request, limiting it initially to a small group of vetted partners — a move that drew criticism for a lack of transparency over how companies are designated as "trusted." Analysts say the Anthropic decision signals the emergence of a more defined regulatory process, though questions remain about how foreign-made models with fewer safety guardrails will be treated in the US market.

The reversal comes amid a turbulent period in Anthropic's relationship with Washington. The company sued the Department of Defense in March after the Pentagon labelled it a "supply-chain risk" over its refusal to allow its AI tools to be used for mass domestic surveillance or fully autonomous weapons. Observers suggest the resolution of the export control dispute reflects efforts by Anthropic leadership to rebuild trust with the administration, though it also raises broader questions about the degree to which the US government may seek to approve or shape the release of cutting-edge AI systems going forward.