Cybersecurity Experts Oppose US Government Ban on Anthropic Models

Leading global cybersecurity experts and industry veterans have addressed an open letter to the US government, demanding the lifting of export restrictions imposed on Anthropic's most powerful AI models — Fable and Mythos. Experts believe such a ban could deprive cyber defenders of critical tools and cause more harm to national security than benefit. This was reported by Techcrunch.com news reports.

The open letter emphasizes that these restrictions have limited the ability to identify and remediate software vulnerabilities. Specialists warn that "while our adversaries are developing rapidly, it is dangerous to unjustifiably strip defenders of the best capabilities." This appeal follows the US government's order to restrict the export of Anthropic models on national security grounds.

Defenders are being disarmed

To date, 76 prestigious experts have signed the letter. Among them are prominent figures such as Alex Stamos, former head of security at Facebook, Jon Callas, former security architecture manager at Apple, and Casey Ellis, founder of the Bugcrowd platform. They fear the government's decision gives an advantage to cybercriminals.

According to Anthropic, when the Mythos model was introduced in April, it was so powerful in finding flaws in security systems that the company was forced to strictly limit access. Initially, only about 50 companies were granted access, which was later expanded to 150 organizations across 15 countries. However, following the government order, Anthropic halted access to these models worldwide.

Questionable reasons behind the ban

The recently released Fable model is a public version of Mythos, featuring strict restrictions to prevent its use in biology, chemistry, and cybersecurity. However, experts argue these restrictions are overly aggressive, blocking even simple security queries. The White House's export control decision may be based on a report suggesting that the protection system in the Fable model could be bypassed (jailbreaking).

According to Katie Moussouris, founder of Luta Security, a confidential report by Amazon researchers demonstrated a method to "break into" the model. However, Moussouris emphasizes that this was not a real breach. Researchers simply asked the model to fix specific errors in open-source code, which she argues is not sufficient grounds for a cyberattack concern.

This situation has reignited debates over the regulation of AI technologies and their role in security. If the US government does not reconsider its decision, domestic cybersecurity companies could lose their competitiveness in the international market, and global networks could remain at risk.