Anthropic has limited access to its new AI model, Claude Mythos Preview, after the system discovered thousands of critical vulnerabilities across operating systems, web browsers, and widely used software. The company warned that AI’s ability to identify and exploit software flaws could soon surpass all but the most skilled human hackers, raising concerns over malicious use.
In 2025, AI powered cyberattacks surged 72% year-over-year, affecting 87% of global organizations, according to AllAboutAI. Anthropic emphasized the risk of these capabilities falling into the wrong hands and announced Project Glasswing, a defensive initiative involving over 40 partners, including Amazon Web Services, Apple, Cisco, Google, JPMorgan, Microsoft, Nvidia, and the Linux Foundation. The project will leverage Claude Mythos Preview to identify vulnerabilities and patch them before malicious actors can exploit them.
Discovery of Long-Standing Software Bugs
Anthropic revealed that many vulnerabilities uncovered are decades old. Highlights include a 27-year-old bug in OpenBSD, a 17-year-old remote code execution flaw in FreeBSD, and a 16-year-old bug in FFmpeg, alongside numerous Linux kernel and cryptography library weaknesses, including TLS, AES-GCM, and SSH. Web applications were also found vulnerable to SQL injection, cross-site scripting, and cross-site request forgery.
AI-Driven Cybersecurity and the Future
While Claude Mythos Preview could dramatically improve software security, Anthropic cautioned that patching vulnerabilities at scale will take years. The company stated that 99% of the flaws identified remain unpatched, underscoring a transitional period where risks remain high. Anthropic envisions a future where AI-driven defenses strengthen global cyber infrastructure, though the path will be challenging.
Disclaimer
This content is for informational purposes only and does not constitute financial, investment, or legal advice. Cryptocurrency trading involves risk and may result in financial loss.