U.S. Lifts Export Controls on Anthropic's Most Powerful AI Models

The U.S. Commerce Department has lifted export controls it imposed on two of Anthropic's most advanced AI models, ending an unusual, weeks-long episode in which the government forced the company to cut off worldwide access on national-security grounds — and reopening a debate over how Washington should govern frontier artificial intelligence.

Simone Bishop July 1, 2026 at 1:48 AM

A short, strange chapter in AI regulation ended this week — one that showed the U.S. government reaching for a tool usually aimed at microchips and pointing it at software.

What changed

The Commerce Department lifted export controls on Anthropic's Fable 5 and Mythos 5 models, and the company said it would begin restoring access, CNBC reported. Commerce Secretary Howard Lutnick said the government had "worked closely with Anthropic" to review the more capable model and "strengthen America's leadership in AI." The move fully reopened access after a partial easing days earlier that had cleared one model for a list of vetted U.S. companies and agencies.

How it started

The controls were imposed in mid-June, days after Anthropic launched the two models, when Commerce ordered the company to immediately suspend access for all foreign nationals — including its own foreign employees — under national-security authorities, Anthropic said. Because the company said it could not quickly separate overseas users from domestic ones across a base of hundreds of millions, the order effectively shut the models down worldwide. Officials' stated concern, according to the reporting, centered on a technique for bypassing the models' safety guardrails that they worried could unlock advanced cyber capabilities — such as finding software vulnerabilities at scale — for adversaries. The Herald is attributing that characterization to the government; it was contested.

Anthropic's objection

Anthropic disputed the rationale. The company said the technique amounted to asking a model to read code and flag flaws — a capability it argued is "widely available from other models," including a competitor's, that were not restricted. Applying such a standard broadly, it warned, would "essentially halt all new model deployments." The company said it disagreed that "a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions."

Critics — and the bigger questions

The government's action drew pushback from outside the company as well. Dozens of cybersecurity researchers signed an open letter urging a more "open, scientific and transparent" process for assessing AI risk, arguing that pulling advanced tools away from defenders while adversaries keep developing their own is counterproductive. A rival AI firm, itself under scrutiny over similar tools, said such government reviews should not become the "long-term default." Analysts drew comparisons to 1990s fights over encryption exports, when controls were eventually loosened as the technology spread regardless, and warned that a temporary ban could hand ground to competitors abroad. The episode leaves the central questions unresolved: what companies must do before releasing frontier models, and whether the government's security-review process will be formalized or remain case by case. (Anthropic is the maker of the Claude AI assistant; the Herald covers it as it would any company.)