Cybersecurity firms are urgently adapting to the rise of Anthropic’s Claude Mythos, an AI model acclaimed for its ability to rapidly identify software vulnerabilities. Its capabilities have attracted intense attention from security companies aiming to understand and mitigate emerging cyber risks. CrowdStrike, a major cybersecurity provider, has been on the frontline, engaging deeply with the model through Anthropic’s Project Glasswing to evaluate its potential impact.
CrowdStrike’s president revealed that demand for access to Mythos surged following news of the company’s involvement in early testing, prompting around-the-clock advisory efforts to help organizations interpret the technology and respond appropriately. The company also participates in OpenAI’s Trusted Access for Cyber program, which grants selective partners early insight into advanced AI tools to explore practical applications and provide feedback on their performance.
With Mythos reputed as “too dangerous to release,” the U.S. government briefly imposed export controls that limited access, although those restrictions have recently been lifted. This regulatory attention signals both the model’s power and its potential risks. CrowdStrike has been actively briefing senior security executives, aiming to dispel misunderstandings and calibrate expectations around Mythos’s capabilities. While the model currently serves primarily as a sophisticated vulnerability scanner, priced at $50 per million output tokens, experts warn that wider availability could enable malicious actors to accelerate exploit discovery, outpacing defenders’ patching efforts.
Despite the initial hype, CrowdStrike reports no major internal surprises, as it already deploys AI models like Opus for vulnerability detection. However, the landscape is evolving quickly. As AI tools become more accessible, the speed and scale of vulnerability discovery may increase dramatically, forcing a rethink of cybersecurity defenses and threat response strategies.