Anthropic has announced its latest artificial intelligence model called Claude Mythos Preview. This new model is the company’s most advanced model to date, outperforming rival models such as the older Claude Opus 4.6 model and Gemini 3.1 on a variety of benchmarks. However, Claude Mythos Preview is so powerful that Anthropic may not be ready to release it yet. The reason is cybersecurity.
The group led by Dario Amodei claims that Claude Mythos is so powerful that it was able to discover thousands of cybersecurity vulnerabilities across apps and operating systems that humans missed.
As a result, Anthropic announced that it is working with more than 40 technology companies to use Mythos to fix vulnerabilities before releasing such models to the public.
Claude Mytos discovered a cybersecurity flaw that humans couldn’t find
The AI startup said in a blog post that Claude Mythos Preview has the potential to identify “thousands of zero-day vulnerabilities” across web browsers and operating systems that can pose significant risks from a cybersecurity perspective. Zero-day vulnerabilities refer to flaws that were not previously identified by software developers.
This included a 27-year-old vulnerability in OpenBSD, an open source operating system commonly used in firewalls and servers known for its security-first approach. Claude Mythos Preview also discovered a number of vulnerabilities in the Linux kernel, which is used to run most of the world’s servers. Anthropic claimed that these flaws could allow an attacker to “escalate from normal user access to complete control of the machine.”
Note that Claude Mythos was first discovered in a data breach of Anthropic’s content management system (CMS). The company acknowledged at the time that this would be its most powerful AI model to date.
Anthropic launches Project Glasswing
Following this observation, Anthropic has decided to postpone, if not cancel, the public release of Claude Mythos. Instead, the company has launched Project Glasswing, which will work with big tech companies to identify and patch such vulnerabilities across a variety of apps and operating systems.
Anthropic CEO Dario Amodei writes about X: “Getting this wrong will obviously be dangerous, but if we get it right, we have a real opportunity to build an internet and world that is fundamentally more secure than it was before the advent of AI-powered cyber capabilities.”
The consortium includes major companies such as Apple, Microsoft, Google, Amazon Web Services, CrowdStrike, Palo Alto Networks, Cisco, Broadcom, and the Linux Foundation.
Anthropic has committed up to $100 million in model usage credits to support Project Glasswing efforts, with partners paying for usage beyond this allocation.
Why is Project Glasswing important?
Anthropic claims that Claude Mythos is just one example of how advanced AI models have become. The company believes these vulnerabilities need to be fixed now before such models become more widespread.
The company’s official X account wrote, “Given the pace of advancement in AI, it won’t be long before models with this much functionality become widespread.”
Note that Claude Mythos has no specific training in cybersecurity, but benefits from his advanced coding and reasoning skills. Anthropic says improved cyber capabilities are a byproduct of its general-purpose design.
Anthropic is working with U.S. government agencies and security organizations to develop safeguards and best practices to responsibly deploy these powerful models.
The startup, led by Dario Amodei, had a run-in with the Pentagon earlier this year after pulling out of a defense pact over disagreements over the unrestricted use of AI.
– end
