In short: After weeks of negotiations over security risks in vulnerability detection, the US government lifts its export ban on Anthropic’s Claude models.
After weeks of negotiations, the US government has lifted the export ban on Anthropic’s AI models Claude 3.5 Sonnet (internally: Mythos 5) and Claude 3.5 Haiku (internally: Fable 5). The models had been completely blocked for more than two weeks because the government feared that security risks in their use for vulnerability detection could be circumvented.
Anthropic announced it would begin restoring access to Claude 3.5 Sonnet and Claude 3.5 Haiku today. The US government had stipulated that only users with a US-based IP address can access these models — a regulation that also applied to Anthropic employees. To meet this requirement, Anthropic blocked both models completely.
Claude 3.5 Sonnet is distinguished by exceptional capabilities in detecting software vulnerabilities, some of which remained undiscovered for decades. US agencies and selected companies use this capability to close security gaps. The government’s central concern was that such capabilities could be misused as a cyber weapon. Claude 3.5 Haiku, released in mid-June, is based on the same technology but has cybersecurity and biotechnology functions blocked. Claude 3.5 Sonnet remains the non-public full version for agencies and authorized companies.
Commerce Secretary Howard Lutnick justified the lifting of the ban by stating that Anthropic has committed to proactively searching for and closing security risks in the models. The trigger for the ban was warnings that the safeguards restricting vulnerability searches could be circumvented — an accusation that Anthropic denied.
The ban had international repercussions: following the US action, OpenAI also restricted access to its latest models to partners approved by the US government. In recent months, Anthropic itself came into conflict with the US government because the company refused to make its AI models available for autonomous weapons systems and mass surveillance in the US. The Pentagon subsequently classified Anthropic as a supply chain risk, which limited the use of Anthropic software in government agencies. Anthropic is suing against this classification. The exceptional capabilities of Claude 3.5 Sonnet for vulnerability detection ultimately made Anthropic indispensable to the US government again.
Source: www.it-daily.net · Published July 1, 2026
Lumi AI News — AI-assisted curation pursuant to Art. 50 EU AI Act. Paraphrase and classification by Lumi News Pipeline v1.7.2.