Anthropic recently unveiled its newest AI model, Claude Mythos Preview, marking a significant advancement over its previous model, Claude Opus 4.6, and competing models like Gemini 3.1. The company has described Claude Mythos Preview as a highly potent AI system, capable of identifying numerous cybersecurity vulnerabilities that eluded human detection across various applications and operating systems.
The AI startup, led by Dario Amodei, highlighted that Claude Mythos Preview uncovered thousands of zero-day vulnerabilities in web browsers and operating systems, posing significant cybersecurity risks. Notably, it identified a 27-year-old vulnerability in OpenBSD and multiple flaws in the Linux kernel, potentially enabling attackers to gain complete control over affected systems.
Initially revealed through a data leak of Anthropic’s content management system, Claude Mythos was anticipated to be the most powerful AI model in the company’s portfolio. However, due to the discovered vulnerabilities, Anthropic has decided to postpone or potentially cancel the public release of Claude Mythos Preview. Instead, the company has initiated Project Glasswing, a collaborative effort with major tech firms to address and rectify these vulnerabilities in diverse applications and systems.
As part of Project Glasswing, Anthropic has partnered with industry giants like Apple, Microsoft, Google, Amazon Web Services, and others. The company has committed up to $100 million in model usage credits to support the project, with participating partners covering additional costs beyond this allocation.
Anthropic underscores the importance of addressing vulnerabilities promptly, emphasizing the need to act decisively before advanced AI models like Claude Mythos become more prevalent. While not specifically tailored for cybersecurity, Claude Mythos leverages its sophisticated coding and reasoning capabilities to enhance cyber defense measures. Anthropic is actively collaborating with government agencies and security organizations to establish guidelines for the responsible deployment of such powerful AI models.
Earlier this year, Anthropic faced a rift with the US Pentagon following disagreements on the unrestricted use of AI technology, leading to the termination of a defense contract. The company remains committed to promoting AI ethics and cybersecurity practices in partnership with various stakeholders.
– Ends