Anthropic just unveiled its most dangerous AI model and is keeping it out of public hands

Project Glasswing unites Apple, Google, Microsoft, and others to fix critical vulnerabilities before the same AI power falls into the wrong hands.

Anthropic has just announced Project Glasswing, which is less of a standard product announcement and more of a global distress flare. The project is bringing a coalition of tech companies together, including Apple, Amazon Web Services (AWS), Broadcom, Cisco, CrowdStrike, Google, Microsoft, NVIDIA, and Palo Alto Networks, to proactively hunt down and patch vulnerabilities in the "world's most critical software infrastructure."

Perhaps the main focus of the announcement is the launch of Claude Mythos Preview, a general-purpose frontier model that Anthropic has deliberately kept unreleased to the general public, and for good reasons. Anthropic revealed that Mythos Preview has autonomously identified thousands of high-severity, zero-day vulnerabilities across all major operating systems and web browsers.

Mythos Preview was able to unearth a remote crash vulnerability in OpenBSD, an operating system famous for its hardened security, sitting unnoticed for 27 years. It also found a critical flaw in FFmpeg, a video processing library, buried in a line of code that automated testing tools had passed over five million times in 16 years. It even autonomously chained together multiple Linux kernel exploits to achieve total machine takeover.

Anthropic isn't new to dominating state-of-the-art frontier models in benchmarks. We already know that the company has been aggressively optimizing its foundation models for complex reasoning and agentic programming, with the launch of models such as Claude Opus 4.5 and Claude Sonnet 4.6 and its massive one-million context window.

However, the exact same agentic reasoning can also be used by a bad actor to automate a zero-day exploit. The barrier to entry for cyberattacks has historically been the human expertise required, but with models like Mythos, that barrier might not exist anymore.

This is why Anthropic is giving the "good guys" a head start by distributing Mythos Preview to over 40 partners and critical infrastructure maintainers. The company is committing up to $100 million in usage credits for Mythos Preview and donating $4 million directly to open-source security entities like the Linux Foundation and the Apache Software Foundation.