Overview
Anthropic is reportedly close to finalising an agreement with the Trump administration that would remove U.S. export restrictions on its two most advanced artificial‑intelligence models, Fable 5 and Mythos 5. The progress, cited by Bloomberg and attributed to Commerce Secretary Howard Lutnick, hinges on the administration’s approval of the security‑related concerns that prompted the original controls.
Background of Restrictions
Two weeks ago the Commerce Department, through a letter from Secretary Lutnick, barred Anthropic from allowing foreign nationals to access Fable 5 and Mythos 5, citing the risk that users could bypass built‑in safety guardrails. In response Anthropic suspended global access to both models while negotiations continued.
Negotiations and Stakeholder Involvement
Senior Anthropic executives, including co‑founder Tom Brown, have met with Lutnick and other officials in recent days. U.S. officials have indicated that the restrictions will be lifted once the government’s security concerns are satisfactorily addressed, though a final decision remains pending approval from the White House and other agencies shaping U.S. AI policy.
Related Developments
The dispute represents one of the most significant U.S. government interventions in a major AI developer’s operations. It follows Anthropic’s confidential filing for an initial public offering at a reported valuation of more than $900 billion. The administration has also broadened scrutiny of advanced AI systems; for example, OpenAI announced it will limit access to a preview of its GPT‑5.6 model to a small group of government‑approved partners.
Model Availability Timeline
Anthropic introduced the Mythos model in April to a limited set of organisations, citing cybersecurity risks. Earlier this month it launched Fable 5 as a public‑facing version with additional safeguards, while Mythos 5 was intended for a smaller group of cybersecurity institutions. The company maintains that completely preventing users from bypassing AI safety protections is not currently possible for any frontier AI developer.