Anthropic will make its new AI model available to select major cybersecurity and software companies in an effort to slow the accelerating AI-driven cyber arms race, the company announced.
Access will be given to firms including Amazon, Apple, Cisco, Google, JPMorgan Chase and Microsoft for cyber defense testing, such as identifying software bugs and simulating real-world attack techniques.
Mythos, officially called Claude Mythos Preview, is not publicly released due to concerns it could be misused by hackers and spies, according to Anthropic. The company has briefed US officials and emphasized the need for safeguards.
Experts say AI systems like Mythos could dramatically accelerate vulnerability discovery, scanning codebases faster than human teams. Anthropic claims the model has already found thousands of previously unknown flaws, though this has not been independently verified. Selected partners also include Nvidia, Broadcom, the Linux Foundation, CrowdStrike, and Palo Alto Networks, aiming to strengthen defenses across critical internet infrastructure.
Officials and researchers warn the cybersecurity landscape is shifting quickly. "If models are going to be this good, and probably much better than this, at all cybersecurity tasks, we need to prepare pretty fast," said Logan Graham. Evron added that while attackers and defenders both have access to AI, defenders must adopt similar tools to keep pace. The broader goal of Anthropic is to deploy Mythos selectively so it strengthens global cybersecurity defenses while limiting exposure to malicious use, reflecting growing concern over AI-powered offensive capabilities across the digital ecosystem.
Industry leaders say the move signals a new phase in cybersecurity, where advanced AI models will be embedded directly into both defensive and offensive workflows. Governments, tech companies, and security vendors are now racing to adapt, as the line between automated protection and automated exploitation continues to blur. Experts expect rapid escalation in AI-driven security competition worldwide in coming years across industries globally expanding.
As AI tools become more integrated into cybersecurity operations, experts emphasize the importance of responsible deployment, rigorous testing, and continuous monitoring. Without such safeguards, both attackers and defenders risk escalating a cycle of faster, more sophisticated cyber operations that outpace traditional security practices across global networks continuously evolving.
Comments