AI is currently being utilized to enhance software security.
Anthropic has introduced a new project named Project Glasswing, partnering with Apple. As part of Glasswing, Anthropic is providing a sneak peek of its recently launched Claude Mythos model to selected collaborators, including Apple.
Anthropic asserts that Mythos has identified “thousands of high-severity vulnerabilities” across “every major operating system and web browser.”
### Apple is included in Anthropic’s Project Glasswing cybersecurity initiative
Apple is featured among the leading technology firms as part of Anthropic’s Project Glasswing group.
> We are announcing Project Glasswing today, a new initiative uniting Amazon Web Services, Anthropic, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks to secure the world’s most vital software.
Moreover, Anthropic mentions that over 40 more entities that “create or maintain critical software” have access to its Mythos Preview AI model. The objective is for these software companies to utilize Mythos to identify and resolve security vulnerabilities before the AI model is made publicly available.
### Claude Mythos has detected thousands of flaws in every major OS and browser
According to Anthropic, Claude Mythos has already been employed to uncover significant security weaknesses in every major operating system and web browser.
> Mythos Preview has identified thousands of high-severity vulnerabilities, including some within *every major operating system and web browser*. Considering the pace of AI advancements, it won’t be long before such capabilities become widespread, possibly even in the hands of those not committed to using them safely. The repercussions—for economies, public safety, and national security—could be considerable. Project Glasswing is an urgent effort to harness these capabilities for defensive ends.
In certain situations, these vulnerabilities have “withstood decades of human oversight and millions of automated security assessments,” the company claims.
For instance, it has identified and linked a cybersecurity vulnerability in the Linux kernel that could grant complete control over a machine.
Cybersecurity proficiency is merely one aspect of strength for the new Claude Mythos AI model. Anthropic’s latest model exhibits improvements over Claude Opus 4.6 in reasoning, agentic search and computer operation, especially in agentic coding.
Anthropic has released a [system card](https://www.anthropic.com/claude-mythos-preview-system-card) outlining the most recent benchmarks for Claude Mythos Preview.
### Availability
“We do not intend to make Claude Mythos Preview widely accessible,” Anthropic states, “but our ultimate aim is to allow our users to deploy Mythos-class models safely at scale—for cybersecurity applications, as well as for the numerous other advantages such highly capable models will offer.”
More information about Project Glasswing and Claude Mythos can be found in Anthropic’s announcement [here](https://www.anthropic.com/glasswing).