Anthropic recently made headlines with its announcement regarding a groundbreaking AI model, Mythos, which has prompted a range of reactions from experts in the field. This new model, described by Anthropic as exceptionally powerful, was initially slated for a wider release. However, the company decided against this, citing significant cybersecurity concerns. Anthropic claimed that the capabilities of Mythos were so advanced that individuals without technical expertise could potentially leverage it to exploit vulnerabilities in major operating systems.
Instead of a public launch, Anthropic has opted to provide access to a controlled preview, known as Claude Mythos Preview, to 11 select organizations, including tech giants such as Google, Microsoft, and Amazon Web Services, as part of its “Project Glasswing.” The decision to limit access, coupled with the company’s dramatic portrayal of Mythos’s capabilities, has stirred considerable debate among AI commentators and industry experts, leading to high-stakes discussions, including a meeting involving federal financial regulators and leaders from major U.S. banks.
Experts’ reactions have varied significantly. Gary Marcus, a noted AI researcher, labeled the hype surrounding Mythos as “overblown.” He contended that while the demonstration did confirm a need for regulatory and technical advancements within the industry, it didn’t present the imminent threat that many in the media portrayed. Marcus characterized Mythos as “incrementally better” than its predecessors rather than a revolutionary breakthrough.
Yann LeCun, founder of AMI Labs and former chief AI scientist at Meta, echoed Marcus’s skepticism. He dismissed the excitement over Mythos, referring to it as “BS from self-delusion,” and noted that similar analyses had been conducted using smaller, less expensive models by other cybersecurity firms.
Conversely, Jake Moore, a global cybersecurity specialist at ESET, acknowledged some elements of marketing language in Anthropic’s announcement but praised the model as potentially impressive and capable of evolving positively over time. He commended Anthropic’s dedication to prioritizing safety in their developments, suggesting that their announcements serve both cautionary and promotional objectives.
Dave Kasten, head of policy at Palisade Research, expressed a belief that while Anthropic’s Mythos may be leading the charge, it might not have a long-lasting competitive advantage. He highlighted insights from a recent Axios report indicating that OpenAI also possesses a similarly advanced model with cybersecurity functions, albeit with limited release plans.
David Sacks, a tech investor and former White House AI czar, urged caution regarding the claims made by Anthropic. He acknowledged the seriousness of the cybersecurity threats associated with Mythos while critiquing the company’s previous tendencies to employ “scare tactics” to enhance their narrative.
The gravity of the situation was underscored by T.J. Marlin, CEO of Guardrail Technologies, who noted that the meeting between federal officials and banking executives aimed to preemptively address potential fallout from a significant cybersecurity breach. He stressed the importance of documentation and readiness among CEOs to mitigate legal risks.
In a contrasting viewpoint, Pablos Holman, a venture capitalist, argued that the advancements in AI could ultimately bolster cybersecurity defenses rather than exacerbate threats. He asserted that defenders in the cybersecurity arena would have access to the same or superior AI resources, enhancing their capabilities to counteract attacks more effectively.
Ben Seri, co-founder of Zafran Security, also acknowledged the potential dangers posed by AI advancements but highlighted that defensive capabilities would need to scale effectively and swiftly to match the evolving threat landscape. He described the current moment as akin to a “Manhattan Project” for cybersecurity, indicating a critical juncture for the industry.
As discussions continue to unfold, the implications of Anthropic’s Mythos and the broader landscape of AI in cybersecurity remain a hotbed of debate among experts, regulators, and the tech industry overall.
