Lead
Anthropic said the US Department of Commerce has withdrawn export controls that had forced the company to suspend access to its most capable Fable and Mythos models, less than three weeks after the order. In a statement on X late on Tuesday the company said, “We’ll begin restoring access tomorrow.” The reversal follows a letter reported by Reuters from US commerce secretary Howard Lutnick saying a licence is no longer required for export. Anthropic also agreed to new monitoring and coordination measures with the US government as part of the resolution.
Key takeaways
- The US Commerce Department lifted export controls on Anthropic’s Fable and Mythos models less than three weeks after ordering a suspension; Anthropic plans to restore access beginning the day after its announcement.
- Anthropic’s most capable versions, Mythos 5 and Fable 5, were disabled after the export-control order on 12 June; Mythos 5 had been partially allowed back to some “trusted” US organisations earlier.
- According to a letter seen by Reuters, Secretary Howard Lutnick said a licence is no longer required and that Anthropic agreed to proactively detect and address security risks tied to the models.
- The US has increased oversight of advanced AI releases amid concerns these systems could be misused by military or intelligence users in countries of concern, including China and Russia.
- The government vetting process has been controversial; OpenAI’s CEO Sam Altman supported safety testing but warned against the government selecting customers.
- Anthropic has committed to work on protocols and standards, to inform the US government of malicious activity, and to collaborate on future releases for Mythos, Fable and successor models.
Background
In recent months US agencies have intensified scrutiny of advanced AI model releases, citing national security risks if powerful models are accessed by adversarial state actors or non-state malicious users. The Biden administration and regulators have signalled that new guardrails may be necessary as foundation models grow more capable and diffuse through private and public sectors. Anthropic, a leading AI developer, emerged as one of several firms whose systems drew specific attention because of their perceived performance and potential dual-use risks.
On 12 June US authorities issued an order that effectively blocked access to Anthropic’s cutting-edge Mythos 5 and Fable 5 models, prompting the company to disable those endpoints. The action formed part of a broader de facto vetting regime in which the US government has evaluated which organisations may be granted access to certain advanced systems. That approach has been criticised by industry leaders for the state’s role in selecting customers while being defended by officials as necessary to prevent misuse.
Main event
The immediate development is a formal withdrawal of the export controls that had restricted distribution of Anthropic’s models. Reuters reported a letter from Commerce Secretary Howard Lutnick stating that the licence requirement had been removed. Anthropic said in a public post that it would begin restoring access the following day, after agreeing to enhanced monitoring and cooperation with US authorities.
Prior to the full lift, the US government partially eased the order by permitting Mythos 5 access for some “trusted” US organisations, a limited roll-out meant to balance security concerns with research and commercial continuity. That partial reversal came the Friday before the full withdrawal and was framed as a cautious, phased approach. Anthropic’s temporary shutdown of Mythos 5 and Fable 5 on 12 June had disrupted partners and customers who had been using the models for testing and product development.
The terms of the accord reported by Reuters say Anthropic will proactively detect and address security risks, work with the US government on protocols and standards for releases, and notify authorities of malicious activity. Company and government representatives framed the agreement as an operational compromise: restoring wider access while instituting stronger safeguards. The announcement did not enumerate every technical requirement or timeline for compliance beyond the near-term restoration pledge.
Analysis & implications
Policy: The commerce action and its reversal illustrate an emergent US posture that mixes export controls, licensing, and cooperative oversight to manage AI risks. By conditioning restored access on company cooperation, regulators aim to set precedents for how commercial AI releases are policed without permanently severing market access. This could become a template for future interventions across other firms and models.
Industry dynamics: The episode signals heightened regulatory leverage over leading AI developers. Firms may face stricter pre-release evaluation, forced telemetry and reporting requirements, and partial roll-outs to vetted partners. While these measures could slow some deployments, companies may adapt by incorporating additional safety workstreams and documentation to satisfy government expectations.
International impact: The explicit concern about misuse by military or intelligence users in countries of concern—cited by US officials—raises the prospect of geopolitical decoupling in advanced AI access. If the US extends similar restrictions to exports, it may spur reciprocal measures from other states or encourage alternative ecosystems that do not depend on US-origin models.
Research and innovation: Mandated monitoring and collaborative protocols could improve detection of misuse and accelerate shared safety standards, but they may also introduce friction for academic and small-enterprise researchers who lack the resources to meet stringent compliance demands. Balancing open research with national security will remain a central policy challenge.
Comparison & data
| Model | Status as of 12 June | Status after recent action |
|---|---|---|
| Mythos 5 | Disabled after export-control order | Partially allowed to trusted US organisations; export controls formally lifted |
| Fable 5 | Disabled after export-control order | Export controls formally lifted; restoration announced |
The table summarizes the operational status changes for Anthropic’s two named model families. The progression — from disablement (12 June) to a phased reintroduction and then a full withdrawal of the licence requirement — illustrates a regulatory path from blunt restriction to conditional re-access. Quantitative details about the number of trusted organisations or the telemetry thresholds required were not disclosed publicly.
Reactions & quotes
Government and industry responses highlighted both relief at restored access and unease about the precedent of government vetting:
“We’ll begin restoring access tomorrow.”
Anthropic statement on X (company statement)
The company framed the move as necessary to resume services while cooperating with safety commitments.
“Anthropic has agreed to proactively detect and address security risks…and to inform the US government of any malicious activity.”
Howard Lutnick, US Commerce Secretary (letter reported by Reuters)
The quoted text, as reported by Reuters, outlines the core conditions that accompanied the withdrawal of the export controls.
“Extensive safety testing is not a bad idea. I just don’t like the idea of the government picking the customers.”
Sam Altman, CEO of OpenAI
Altman’s comment echoes wider industry concerns about the balance between safety regimes and market access decisions made by government agencies.
Unconfirmed
- Whether any foreign state actors had gained access to Mythos or Fable prior to the June 12 order remains unconfirmed by public records or government statements.
- The exact technical criteria and telemetry thresholds Anthropic must meet to satisfy US authorities were not disclosed and remain unconfirmed.
Bottom line
The Commerce Department’s withdrawal of export controls restores broader access to Anthropic’s Fable and Mythos models while making clear that future releases will be subject to ongoing scrutiny and cooperative safeguards. For Anthropic, the agreement avoids a prolonged market cutoff but introduces continuing compliance obligations that will shape future product roadmaps.
More broadly, the episode marks a tactical model for US regulators who seek to manage AI risk without permanently severing commercial ecosystems. Stakeholders should watch for published protocols, the scope of telemetry sharing, and whether the approach is replicated for other companies and model families.
Sources
- The Guardian — news report summarising the development and context
- Reuters — news agency report (letter to Anthropic seen by Reuters)
- Agence France-Presse (AFP) — news agency reporting