- Former Facebook CSO Alex Stamos warned that restrictions on advanced AI models could weaken cyber defenses as AI-powered threats continue to grow.
- More than 40 cybersecurity leaders, including executives from Adobe, Zoom, Sophos, and Nvidia, urged the Trump administration to reverse limits on Anthropic’s Mythos AI models.
- Security experts argue that advanced AI tools help defenders identify software vulnerabilities, generate proof-of-concept code, and accelerate threat detection.
- Critics of the restrictions say similar capabilities already exist in competing AI models, including OpenAI’s GPT-5.5, Anthropic’s Opus and Sonnet, and Chinese AI models such as Kimi 2.7.
According to a report by Axios, former Facebook Chief Security Officer Alex Stamos warned that new restrictions on advanced artificial intelligence models could weaken cybersecurity efforts at a time when AI-powered threats are increasing.
Axios, a U.S.-based digital news outlet, reported that leading figures in cybersecurity, including CISOs and executives from Adobe, Zoom, and Sophos, are pleading with the Trump administration to lift limits on Anthropic’s most advanced AI models, claiming that doing so would harm cyber defenders more than attackers.
The leaders contend that restricting access to Anthropic’s first publicly accessible Mythos-class model might cripple cyber defenses just as they’re preparing for a surge of AI-powered hacking threats.
The loosely organized group of experts, led by former Facebook Chief Security Officer Alex Stamos, contend in the letter that other top AI models also have the problem that Amazon researchers identified.
More than 40 people had signed the letter as of Sunday night. The letter was signed by Katie Moussouris, CEO of Luta Security, Rachel Tobac, CEO of SocialProof Security; Chris Wysopal, co-founder of Veracode, Paul Vixie, a renowned computer scientist, Joe Levy, CEO of Sophos; and Aaron Grattafiori, security researcher at Nvidia.
AI Restrictions Raise Cybersecurity Concerns
According to the letter, “this action has taken the best models away from defenders, created market uncertainty, and risked America’s AI leadership without any real risk to justify it.”
Cyber experts warn Fable limits aid attackers and hurts defenders https://t.co/kZFhgLjq0M
— Axios (@axios) June 15, 2026
Anthropic referred Axios to its Friday statement regarding the directive from the U.S. government when asked for a statement. A business representative told Axios on Friday that Amazon does not disclose specifics of cybersecurity talks it has with countries, but the company did not immediately reply to a request for comment.
Stamos, who is currently Corridor’s chief product officer, told Axios that the Fable 5 security features that seemed to worry the White House were the model’s capacity to produce a “proof of concept” for vulnerabilities.
According to him, the “proofs of concept” can help security teams learn how to safeguard their systems, but they can also produce a blueprint for code that allows antagonistic hackers to access a system.
Fable 5 Sparks Cybersecurity Debate
According to Stamos, only Mythos 5 and Mythos Preview, which were previously exclusively accessible to approved participants in Anthropic’s Project Glasswing effort, could transform those proofs of concept into completely independent attack chains.
“You cannot give Fable the entire Linux kernel and say ‘Find all the security bugs,'” Axios was told by Stamos. He continued, “Amazon is not claiming that they can do this.”
Cybersecurity experts immediately questioned Anthropic’s stringent safeguards when it introduced its Fable 5 model last week. The model’s shortcomings became a hot topic in the cybersecurity industry, according to an open letter signed by security leaders.
The signatories claimed that other top AI systems, such as OpenAI’s GPT-5.5, Anthropic’s Opus and Sonnet models, and Chinese AI models like Kimi 2.7, already have comparable capabilities.
Chinese AI Models Narrow The Gap
Alex Stamos, a former CSO at Facebook, said that open-source Chinese methods are quickly catching up in terms of software vulnerability detection and analysis.
Stamos claimed that limiting access to powerful AI capabilities in the United States would put cyber defenses at a disadvantage while global competitors continued to develop similar technologies.
“We are in a race right now to fix these bugs as fast as possible,” Stamos stated. Limiting access to cutting-edge AI capabilities while attackers continue to find and accumulate vulnerabilities might undermine overall cybersecurity efforts, he continued.
You need to login in order to Like
Leave a comment