Anthropic Restricts Mythos AI Over Cybersecurity Weaponization Risks

Anthropic has unveiled a powerful new AI technology called Mythos, capable of identifying thousands of software vulnerabilities across major operating systems and web browsers. Due to the potential for the tool to be weaponized by cybercriminals, the company has opted against a public release, fearing the fallout for national security, public safety, and global economies could be severe.

A Targeted Defense Strategy

Rather than making the model widely available, Anthropic is launching Project Glasswing. This initiative provides select major companies—including Apple, Amazon, Cisco, JPMorgan Chase, and Nvidia—access to Mythos. By allowing these organizations to test the model, Anthropic aims to help them harden their IT infrastructure against potential cyberattacks before the technology or similar AI models fall into the hands of bad actors.
Security experts view this development as a stark warning regarding the speed and capability of AI-driven threats. Alissa Valentina Knight, CEO of the cybersecurity firm Assail, noted that the industry is struggling to keep pace with adversaries who are increasingly leveraging AI to execute attacks with greater speed and efficiency than human hackers ever could.

Rising Risks to Global Systems

The emergence of Mythos has drawn significant attention from federal officials and international leaders. Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell recently held a closed-door meeting with top bank CEOs to discuss the cybersecurity risks posed by AI. Furthermore, IMF Managing Director Kristalina Georgieva has expressed concern regarding the world's ability to protect the international monetary system, noting that cyber risks are growing exponentially.
The threat is compounded by the fact that hackers are already utilizing advanced AI to create autonomous agents capable of spreading malware, executing identity theft, and launching ransomware attacks. According to a report by PwC, the time between the public release of a new AI capability and its weaponization by threat actors has shrunk dramatically, a trend expected to accelerate throughout 2026.

Balancing Safety and Marketing

While Anthropic’s cautious approach aligns with its established focus on AI safety, some industry observers suggest the strategy may serve dual purposes. Peter Garraghan, founder of the AI security platform Mindgard, speculated that the limited release could function as a marketing tool, particularly as reports suggest Anthropic may be preparing for an initial public offering by the end of the year.
Marketing experts, however, point out that the move reinforces the company’s brand identity. By curating the release of Mythos, Anthropic is positioning itself as a leader in responsible AI development. Regardless of the underlying motivations, the consensus among security professionals remains clear: as AI becomes more adept at scanning code for bugs that humans often miss, the frequency of cyberattacks is likely to increase until organizations can patch vulnerabilities in near real-time.