21 days · 8 summary articles
Nearly 150 of the United States’ most prominent cybersecurity researchers warned on Monday that the Trump administration’s clampdown on Anthropic’s advanced AI models risks crippling America’s defensive cyber capabilities and handing adversaries an enduring advantage. In an open letter organised by former Facebook security chief Alex Stamos, the signatories urged the White House to reverse its decision to restrict access to Anthropic’s Fable 5 and Mythos 5 models, arguing that the move sets a dangerous precedent that could deter U.S. AI firms from building tools that help defenders identify and fix vulnerabilities.
The dispute erupted last week after Anthropic publicly released Mythos 5, prompting concerns from Amazon CEO Andy Jassy that the model could be exploited by Chinese actors. Anthropic brought in a leading zero-day bug hunter—who helped create the Pentagon’s bug bounty programme and has served on multiple government advisory boards—to assess Amazon’s security claims. Yet the administration has since labelled the researcher a “radical Democrat,” according to Axios reporting .
Security leaders say the administration’s response misrepresents the actual risk. Katie Moussouris, CEO of Luta Security, noted in a blog post that Amazon’s concerns centred on a narrow jailbreak allowing Fable 5 to generate “proofs of concept”—a capability defenders routinely use to understand and mitigate vulnerabilities. “There is no fix that would not render the model less useful for cyber defenders,” Moussouris wrote on X. “No new frontier models can be developed or released if this is the administration’s best take.”
The broader implications are stark. Cybersecurity experts warn that if U.S. AI companies fear punishment for models capable of identifying flaws, they may strip out critical defensive features, leaving American systems more exposed. Stamos, who has consulted with technical teams involved in the dispute, told Axios that the vulnerabilities Amazon flagged are not unique to Anthropic’s models. “They’ve set a precedent that American models can’t do defensive security research,” he said.
Researchers also warn that adversaries like China are unlikely to abandon similar tools, potentially widening the capability gap between U.S. defenders and their rivals. “This is closer to China than what I recognise as the United States,” Stamos said, calling the policy “a huge threat to American dynamism.”
The administration is meanwhile standing up a vulnerability clearinghouse under its recent AI security executive order, which would triage reports of jailbreaks and prompt injections. Yet questions persist about the depth of cybersecurity talent remaining in the White House after recent departures and the sidelining of the nation’s top cyber agency. With the G7 summit in Évian this week, the Anthropic dispute has become a flashpoint over whether the U.S. is prioritising short-term security theatre over long-term technological resilience.
Follow us for live European news
2 further sources not geolocated