Anthropic has raised alarms about the implications of its latest AI model, Claude Mythos Preview, on cybersecurity. The model is capable of analyzing large software codebases to find thousands of “zero-day” vulnerabilities, which could significantly speed up cyberattacks, potentially reducing the timeframe for planning and executing such attacks from months to mere hours or minutes.
Launched in 2026, Project Glasswing involves over 40 tech companies, including Microsoft, Google, and AWS, aiming to enhance cybersecurity. Access to Claude Mythos Preview is currently restricted to mitigate risks associated with its potential misuse. The firm emphasizes the need for improved cybersecurity measures, arguing that AI's automated capabilities could enable even inexperienced individuals to execute complex cyberattacks.
Anthropic is advocating for updated cybersecurity protocols and clearer regulations to combat AI-driven threats, highlighting that human oversight remains vital despite advancements in technology. The company calls for intensified collaboration among tech firms, government bodies, and researchers to address these emerging challenges effectively.