Cybersecurity vets protest ‘dangerous’ US government ban on Anthropic’s most powerful models

A coalition of cybersecurity professionals, featuring notable industry veterans, has sent an open letter to the U.S. government urging the removal of export restrictions on Anthropic's advanced AI models, Fable and Mythos. The letter argues that this ban hampers cybersecurity efforts, stating, "This action has taken the best models away from defenders who rely on them to identify vulnerabilities and enhance the security of their products." The experts expressed concern over the risks posed by restricting access to such powerful tools, especially as cyber adversaries continue to evolve. On Friday, the U.S. government mandated that Anthropic limit the export of these models due to national security apprehensions, but the specific motivations behind this decision remain unclear, as noted by Anthropic. Following the government's order, Anthropic halted global access to both Fable and Mythos. The letter has garnered support from 76 cybersecurity specialists, including former Facebook security chief Alex Stamos and Bugcrowd founder Casey Ellis. Other signatories include renowned cryptographer Jon Callas, computer scientist Paul Vixie, and Rachel Tobac, CEO of SocialProof Security. Anthropic previously highlighted the remarkable capability of Mythos in identifying security vulnerabilities when it previewed the model in April. Due to its potential misuse by malicious actors, access was initially restricted to about 50 companies, later expanding to 150 organizations across 15 countries. Last week, Anthropic introduced Fable, a public iteration of Mythos, which features strict limitations aimed at preventing its utilization in fields like cybersecurity. However, many cybersecurity professionals found that Fable’s guardrails were so stringent that they significantly restricted discussions related to cybersecurity itself. The need for these limitations reportedly arose from a report suggesting a method to bypass Fable's controls, allowing access to its more powerful capabilities akin to Mythos. Katie Moussouris, a signatory of the letter, commented on the potential jailbreak method mentioned in a non-public Amazon research paper. According to her assessment, the paper did not convincingly demonstrate a legitimate jailbreak. Instead, it involved prompting Fable to correct open-source code with known vulnerabilities, which Moussouris argued is an essential function for enhancing defense against cyber threats. The open letter emphasizes the necessity for transparent and fair regulations developed through a democratic process, grounded in scientific research. The experts advocate for these regulations to be applied only as needed to ensure public safety, stressing the importance of empowering defenders with optimal tools for their cybersecurity efforts.