US Lifts Curbs on Anthropic's Advanced AI Models Amid Tech Race

U.S. AI developer Anthropic has reinstated access to its advanced Fable 5 and Mythos 5 models, following the U.S. Commerce Department's decision to remove export restrictions imposed three weeks prior. This development occurs amidst a heightened race between American and Chinese artificial intelligence developers.

Fable 5 was initially pulled after Amazon researchers highlighted a method to circumvent its security safeguards. U.S. authorities deemed this vulnerability a potential national security risk, leading to restricted access for international users. Anthropic publicly disagreed with the decision, stating that similar vulnerabilities were detected internally in other advanced models, including their Opus 4.8 and GPT-5.5.

“

We have detected similar vulnerabilities in other advanced models, including our Opus 4.8 and GPT-5.5.

Some cybersecurity experts, however, considered the initial vulnerability's risk to be overstated. Fable 5 returns with modifications; Anthropic has addressed the bypass method, though this impacts certain programming and debugging tasks. Some of these requests will now be rerouted to the Opus 4.8 model. For subscribers, Fable 5 will be available at half its weekly quota until July 7, after which it will be billed based on usage, at approximately double the cost of Opus 4.8.

“

The lifting of restrictions is not only important for Anthropic but also for the broader U.S. strategy in artificial intelligence.

The reinstatement coincides with escalating competition from China. U.S. media reports indicate that Chinese AI models have achieved results comparable to Anthropic's Mythos 5 in certain cybersecurity tests, while being significantly cheaper. The New York Times' Hard Fork podcast discussed how the lifting of restrictions is crucial not only for Anthropic but also for the broader U.S. AI strategy, emphasizing the delicate balance between national security, regulation, and maintaining technological advantage over China. The podcast also anticipates a similar resolution for OpenAI's upcoming GPT 5.6 model, which is also reportedly under temporary export restrictions.

For Anthropic, regaining global access to its models is also a business imperative as the company accelerates revenue growth ahead of its planned initial public offering (IPO).

“

The case shows how delicate the balance is becoming between national security, regulation, and maintaining technological advantage over China.