Anthropic sends researcher to address US government AI safety concerns

Anthropic, the company behind the Claude family of AI models, sent one of its top security researchers to Washington to address escalating government concerns about the safety of frontier AI systems. The move comes as US export control directives forced the company to temporarily pull global access to two of its newest models.

Export controls hit Anthropic’s newest models

In June 2026, US export control directives led Anthropic to temporarily suspend global access to its Fable 5 and Mythos 5 models. The reasoning: vulnerabilities that could allow adversaries to jailbreak the software, effectively stripping away cybersecurity protections and opening the door to malicious uses.

The suspension of model access is notable because Anthropic has long positioned itself as the safety-first alternative in the AI race. The company signed memorandums of understanding with the US AI Safety Institute back in 2024, committing to pre-release model safety evaluations.

Carlini’s role at Anthropic focuses specifically on testing how well model defenses hold up against adversarial attacks. Red-teaming, his specialty, involves deliberately trying to break a system to find its weak points before someone with worse intentions does.

Tensions with the Trump administration

The researcher’s visit to Washington unfolds against a backdrop of growing friction between Anthropic and the Trump administration over how AI technologies should be regulated, particularly where those capabilities intersect with national security.

Tensions escalated in 2026 over disputed Pentagon policies regarding AI usage, with the administration raising concerns about the capabilities of Anthropic’s models. Anthropic has been actively engaging with US officials on multiple fronts, deploying not just Carlini but a broader team of technical experts and elite security engineers to navigate what has become an increasingly complex regulatory environment.

The proliferation of jailbreak vulnerabilities has fueled concern in Washington about what unrestricted access to frontier AI models could mean in cyber warfare scenarios. Legislative responses have focused on striking a balance between fostering innovation and imposing rigorous safeguards.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.