Aug 24, 2026 12:44:00

AI company Anthropic has announced that its highly capable AI model ‘Claude Mythos 5’ in the cybersecurity field is now available in its security product ‘Claude Security.’ Previously, its use was restricted due to concerns about the risk of misuse in attacks, but the company has mitigated this risk by using Mythos 5 only for predetermined processes in the background and by creating a system that only receives predetermined results such as vulnerability fixes and security warnings.

Bringing the cybersecurity capabilities of Claude Mythos 5 to more defenders | Claude by Anthropic

https://claude.com/blog/bringing-claude-mythos-5-to-more-defenders

As the coding capabilities of generative AI improve, its ability to find and fix software vulnerabilities also increases. However, the same capabilities can also be used to create attack code that exploits those vulnerabilities, creating a difficult problem in cybersecurity: ‘providing powerful AI to the defenders while preventing attackers from using it.’

In April 2026, Anthropic launched ‘Project Glasswing,’ providing Claude Mythos Preview to a select group of organizations protecting critical software and infrastructure. According to Anthropic, Mythos Preview discovered thousands of high-severity vulnerabilities in software, including major operating systems and web browsers, with the aim of giving defenders time to discover and fix vulnerabilities before similar AI solutions become widely available. The successor model, Claude Mythos 5, was also initially provided to a limited number of organizations.

Anthropic develops ‘Claude Mythos Preview,’ an AI with extremely high cyberattack capabilities, and has also launched ‘Project Glasswing,’ which will provide a preview version to Microsoft, Apple, and others – GIGAZINE

On the other hand, limiting the organizations that can use it would prevent Mythos 5’s defensive capabilities from reaching a wide range of security professionals. Anthropic has made ‘Claude Fable 5,’ which uses the same underlying model as Mythos 5, generally available, but Fable 5 incorporates security measures that restrict cybersecurity-related requirements, preventing it from providing full defensive capabilities.

The official version of ‘Claude Mythos’ has finally been released, and ‘Claude Fable,’ a version with no user restrictions, has also been released, making it available to everyone – GIGAZINE

If users can freely input prompts, it creates an opportunity for malicious users to manipulate instructions to generate attack code that exploits vulnerabilities. Therefore, by using Mythos 5 only for predetermined processes in the background and returning only the necessary results to the end user, it has become possible to safely integrate Mythos 5 into security products. For example, in a product that fixes vulnerabilities, the user will receive a ‘list of recommended fixes.’ Anthropic and its partners are also implementing mechanisms to prevent misuse so that the model does not deviate from its intended use.

In the case of Anthropic’s vulnerability analysis service ‘Claude Security,’ which is available as a public beta for Claude Enterprise users, Mythos 5 examines the source code of a specified GitHub repository to search for vulnerabilities. The findings include a Common Weakness Enumeration (CWE) that categorizes software weaknesses, as well as the confidence level, severity level, and recommended remediation actions. While proposed remediations can be implemented using the web version of Claude Code, the actual interactive remediation process uses models available to the organization within Claude Code, rather than Mythos 5.

Along with expanding the use of Mythos 5, Anthropic also announced the ‘Defender Advantage Fund (0xDAF)’ to support improvements in the security of open-source software. It plans to provide Claude usage credits worth a total of $35 million (approximately 5.6 billion yen) to support vulnerability fixes in widely used open-source software and the automation of vulnerability detection and remediation work.

Furthermore, for the ‘Cyber Verification Program’ for security organizations that have passed the screening process, restrictions related to cybersecurity will be eased for Claude Opus and Claude Sonnet, and in the future, defensive capabilities using Mythos-class models will also be made available. Anthropic says it will work with government agencies, corporations, and open-source software developers to develop security measures and access systems, and provide powerful AI-powered defensive capabilities to more organizations.