What smart people are saying about Mythos, Anthropic's new AI model that has some cybersecurity experts spooked

Anthropic's recent unveiling of its advanced AI model, Mythos, has ignited a flurry of warnings and skepticism among cybersecurity experts. The company announced that it would not release Mythos widely due to security concerns. According to Anthropic, the model's capabilities could potentially be wielded by non-experts to exploit vulnerabilities in significant operating systems. Instead of a broad launch, Anthropic has opted to provide a preview of Claude Mythos to a select group of 11 organizations, including tech giants like Google, Microsoft, and Amazon Web Services, under the initiative known as Project Glasswing. This cautious approach has raised eyebrows, particularly as it coincided with a high-level meeting involving Federal Reserve Chair Jerome Powell, Treasury Secretary Scott Bessent, and leaders of major U.S. banks. While some AI experts expressed alarm over the implications of Mythos, others questioned its groundbreaking status, suggesting it may not be significantly ahead of its competitors. Gary Marcus, a prominent AI researcher, characterized the announcement as "overblown," asserting that the model seemed only "incrementally better" than existing technologies, rather than a revolutionary advancement. Yann LeCun, founder of AMI Labs, echoed this sentiment, dismissing the hype surrounding Mythos as exaggerated. He noted that smaller and less expensive models had demonstrated similar capabilities regarding the vulnerabilities highlighted by Anthropic. On the other hand, Jake Moore, a global cybersecurity specialist at ESET, acknowledged that while some marketing language was present in Anthropic's announcement, the model itself remains impressive and is expected to evolve positively. He emphasized that the company has established itself as a leader in safety-focused AI, which lends credibility to its cautious approach. Dave Kasten from Palisade Research suggested that other AI models may soon catch up to Mythos. He indicated that while Anthropic might have a slight edge, the competitive landscape is rapidly shifting, especially with reports of OpenAI developing a similarly advanced model. David Sacks, a tech investor, cautioned that while the potential cybersecurity threats posed by Mythos should be taken seriously, it is essential to remain aware of Anthropic's history of employing scare tactics. The meeting among federal officials and Wall Street executives was framed as a precautionary measure to ensure that banks are prepared for potential security breaches. T.J. Marlin, CEO of Guardrail Technologies, pointed out that failing to document a board-level response could leave CEOs in a precarious legal position. In a contrasting perspective, Pablos Holman, a VC at Deep Future, argued that defenders in cybersecurity are likely to benefit more from advancements in AI than attackers, highlighting that both sides now have access to powerful tools. Ben Seri, co-founder of Zafran Security, described the current moment in cybersecurity as akin to a "Manhattan Project," emphasizing the urgent need for defenders to adapt quickly. He warned that while AI can expedite the identification and resolution of vulnerabilities, the real challenge lies in the safe and swift implementation of fixes in live environments. As the conversation around Mythos continues, the balance between innovation and safety within AI technologies remains a critical focal point for experts and industry leaders alike.