Anthropic's Mythos AI System Might Actually Create More Cybersecurity Vulnerabilities

Anthropic, the company behind the popular AI assistant Claude, is now testing an unreleased version of its frontier AI model: Claude Mythos Preview. It's currently only available to select tech firms, but findings have shown that it has the potential to create numerous cybersecurity vulnerabilities when it is released to the world. According to Anthropic, "Mythos Preview has already found thousands of high-severity vulnerabilities, including some in every major operating system and web browser." This is unsettling news in the wake of a recent Google report discussing AI as a hacker super weapon.

However, Anthropic's Mythos AI won't see the light of day until the company is more confident that the world is ready for the cybersecurity risks it might present. As part of an initiative called Project Glasswing, Anthropic is partnering with many of the most influential tech companies "in an effort to secure the world's most critical software." Participants in Project Glasswing include Apple, Amazon Web Services, Google, Microsoft, and other well-known names.

Mythos is proving to be so good at coding that it can easily identify undiscovered zero-day vulnerabilities in codebases that were thought to be secure. Anthropic's partners in Project Glasswing are currently using Mythos Preview for defensive purposes, while experts continue their research and share their knowledge across the industry. Even so, there is a very real possibility that cybercriminals will soon be using Claude Mythos and similarly powerful models to attempt unprecedented feats of hacking in the near future.

The purpose of Claude Mythos is to advance AI capabilities in areas that include software engineering, reasoning, and research. Anthropic (link downloads a PDF to your device) claims that Mythos Preview demonstrates a "striking leap in scores on many evaluation benchmarks compared to our previous frontier model, Claude Opus 4.6." This huge leap is exactly what makes Mythos such an ominous force.

Recently, a Chinese supercomputer suffered the biggest hack in the country's history and there are new hacking threats that steal passwords through 2FA. If it ever seems like AI technology is advancing too fast, keep in mind that hacking tools are developing just as fast, if not faster. Nothing is stopping malicious parties from using the best AI models for their own purposes, either. Some voices are more optimistic in the face of AI advancement.

In an April earnings call (via CNBC), JPMorganChase CEO Jamie Dimon said, "[AI] does create additional vulnerabilities, and maybe down the road, better ways to strengthen yourself too." JPMorganChase is one of the organizations deploying Mythos Preview as part of Project Glasswing. The company is invested in staying ahead of the curve on cyber threats; a philosophy that will ideally pan out for Project Glasswing as a whole.