Cybersecurity Experts Demand Reversal of Ban on Anthropic Models

A group of dozens of cybersecurity experts has appealed to the White House to lift export restrictions on Anthropic's flagship models, Fable and Mythos. The authors call the ban "dangerous" and warn that it constrains precisely those who work to protect software and corporate infrastructure.

What the letter says

The letter was signed by dozens of industry veterans, specialists with years of experience from private companies and academic institutions. Their key argument: the attacking side, meaning hackers, cybercriminals and state-backed APT groups, is not bound by regulatory barriers. Attackers continue to freely use powerful AI models to discover vulnerabilities, write exploits and automate attacks at industrial scale.

Defenders, meanwhile, are put in an artificially losing position: security specialists outside permitted jurisdictions are deprived of the same tools and work with demonstrably fewer capabilities.

What defenders stand to lose

Fable and Mythos, Anthropic's flagship models, perform strongly in tasks that are critical to cybersecurity: analyzing code for vulnerabilities, interpreting complex binaries and synthesizing technical threat reports. That makes them valuable above all for defensive work.

Specific uses that are affected:

• Automated code auditing and detection of zero-day vulnerabilities
• Generation and verification of security patches
• Malware analysis and reverse engineering
• Threat intelligence automation and incident classification
• Development and updating of security playbooks for response teams

The more complex the threat and the more sophisticated the code, the more the quality of the AI assistant matters. That is why restricting access to the most powerful models hits analysts first, not attackers.

The regulatory logic and its weak points

Export restrictions on frontier models are part of a broader policy: US regulators treat powerful AI systems as a strategic resource comparable to semiconductor technology. The stated goal is to keep competing states from using advanced systems for military or intelligence purposes.

The letter's authors point to a fundamental contradiction in this logic. The most dangerous uses of AI, such as automating cyberattacks, developing exploits and creating deepfakes, run through open models and workaround channels. No export control closes those off. The defensive community, however, including MSSPs, independent researchers, corporate SOCs and bug bounty teams, takes direct operational losses.

Cybersecurity is global by nature: threats do not recognize borders, and effective defense requires international cooperation. Restrictions that cut foreign partners and allies off from the best tools weaken the entire ecosystem's response to threats, including those aimed at the US itself.

What this means

The veterans' letter is one of the first coordinated signals from the industry against a new generation of AI export restrictions. It forces regulators to choose: keep blocking powerful models wholesale, or move to a more targeted approach that restricts specific uses rather than the tools themselves. Until that distinction exists, the control works as a barrier for defenders, but not for attackers.