Anthropic is pushing further into AI-powered cybersecurity with Claude Mythos 5, its most capable model for cybersecurity and biology research. The company says Mythos 5 is the latest update to its Mythos-class models and has been made available to a small group of trusted testing partners because of its advanced cyber capabilities. At the same time, Anthropic is expanding its Claude Security offering, which uses its latest public frontier models to scan codebases and suggest security patches.
Claude Mythos 5: Anthropic’s Cybersecurity-Focused Model
Anthropic introduced Claude Mythos 5 on June 9, 2026, describing it as an updated version of the earlier Claude Mythos Preview. According to the company, Mythos 5 delivers improvements across cybersecurity, biology and healthcare benchmarks. Unlike Anthropic’s general-purpose models, Mythos 5 is currently available only to a limited group of initial testing partners for cybersecurity and research applications.
Anthropic has placed additional restrictions around access because highly capable cybersecurity models can be useful to defenders but can also potentially be misused for offensive purposes. The company therefore uses a trusted-access approach for Mythos 5 rather than making its unrestricted cyber capabilities broadly available.
Claude Security Brings AI Into Enterprise Vulnerability Scanning
Alongside Mythos 5, Anthropic has developed Claude Security, a product aimed at helping security teams analyze software for vulnerabilities.
Anthropic says Claude Security uses its latest publicly available frontier models, including Claude Opus 4.8, to scan codebases and suggest patches. This is an important distinction: Claude Security should not be described as simply being powered by Mythos 5. Mythos 5 and Claude Security are related to Anthropic’s broader cybersecurity effort, but they are separate offerings with different access models.
The product is intended to help security teams identify weaknesses in software and move from vulnerability discovery toward remediation.
Project Glasswing and the Earlier Mythos Preview
Anthropic’s cybersecurity work began publicly with Project Glasswing, launched in April 2026. The initiative brought Anthropic together with technology companies, security organizations, researchers and critical software providers to use advanced AI models to help secure important software.
The initial program used Claude Mythos Preview, the predecessor to Mythos 5. Anthropic said that its approximately 50 initial partners had used the model to scan codebases and had identified more than 10,000 high- or critical-severity vulnerabilities. Anthropic subsequently expanded the program to approximately 150 additional organizations across more than 15 countries.
This makes Project Glasswing an important part of the story behind Anthropic’s increasingly aggressive focus on AI-assisted cybersecurity.
Mythos Models Can Find Previously Unknown Vulnerabilities
Anthropic’s research indicates that Mythos-class models are capable of much more than traditional automated code checking.
In its assessment of Claude Mythos Preview, Anthropic reported that the model was able to identify and exploit previously unknown vulnerabilities, including vulnerabilities in major operating systems and web browsers, when directed to do so under authorized testing arrangements. The company also reported that the model could discover application-logic vulnerabilities and vulnerabilities in low-level software.
Anthropic has emphasized that many of these findings have not been publicly disclosed because they were still undergoing coordinated vulnerability disclosure and remediation processes.
That is one reason the company has been cautious about broadly releasing the most powerful cyber capabilities of its models.
AI Could Change How Vulnerability Research Is Done
Traditional cybersecurity research can require security engineers to spend significant amounts of time reviewing source code, understanding software architecture, tracing execution paths and reproducing potential vulnerabilities.
Advanced AI models can potentially automate portions of this process.
An AI system can examine large codebases, reason about how different components interact and investigate suspicious behavior. When combined with an agentic workflow, the model can also perform multiple steps in sequence rather than simply answering a developer’s question.
Anthropic’s research on Mythos Preview suggests that this capability is becoming increasingly significant. The company says the model can autonomously discover vulnerabilities and, in some cases, develop proof-of-concept exploits under authorized testing conditions.
Claude Security Focuses on the Defensive Side
The enterprise opportunity is particularly interesting because organizations do not necessarily need direct access to an unrestricted cybersecurity model to benefit from AI.
Claude Security is designed to package advanced model capabilities into a workflow for security teams. Instead of requiring every security professional to interact directly with a highly capable cyber model, the product can analyze code and produce security findings and suggested fixes.
Anthropic says its Claude Security product can scan codebases and suggest patches, bringing AI-assisted vulnerability discovery and remediation into a more practical enterprise workflow.
This approach could make AI-assisted security more accessible to companies that have large software portfolios but limited cybersecurity resources.
Mythos 5 Is Not the Same as Claude Security
This distinction is important for understanding Anthropic’s announcement.
Claude Mythos 5 is a highly capable cybersecurity-focused model with restricted access for trusted testing partners.
Claude Security is an enterprise security product that Anthropic says uses public frontier models such as Claude Opus 4.8 to scan codebases and suggest patches.
Project Glasswing is Anthropic’s broader collaborative cybersecurity initiative, which initially used Claude Mythos Preview with selected partners.
Therefore, it would be inaccurate to say simply that “Claude Security is powered by Claude Mythos 5.” The three developments are connected, but they should be presented separately.
The Growing Cybersecurity Challenge of AI
Anthropic’s work also highlights a broader issue for the cybersecurity industry.
As AI becomes better at discovering vulnerabilities, defenders can use the same capabilities to find and fix weaknesses before attackers exploit them. However, increasingly capable models could also lower the technical barrier for malicious actors.
Anthropic itself has acknowledged this dual-use risk. Its Mythos 5 documentation says the model’s advanced cybersecurity capabilities could potentially be used for both defensive and harmful purposes, which is why access remains restricted.
This creates a race between AI-assisted defense and AI-assisted attacks.
Anthropic’s AI Cybersecurity Strategy
Anthropic’s recent developments show a multi-layered approach to cybersecurity.
At the highest capability level is Mythos 5, available to a limited set of trusted partners. Project Glasswing provides a collaborative environment for using advanced AI to protect important software. Meanwhile, Claude Security brings AI-powered vulnerability scanning and patch suggestions into an enterprise-oriented product using publicly available frontier models.
Anthropic has also continued researching AI-assisted cybersecurity beyond software vulnerability discovery. In July 2026, the company reported that researchers using Claude Mythos Preview discovered new weaknesses involving cryptographic algorithms, although Anthropic said the findings did not currently affect production systems.
What This Means for Developers and Security Teams
For software developers, the growing capabilities of AI security tools could mean that vulnerability checking becomes increasingly integrated into the development process.
Instead of waiting for a security audit after software is built, AI-assisted tools could continuously examine code during development, identify potential problems and help developers understand how to fix them.
For security teams, the biggest potential benefit is scale. AI can examine large amounts of software and help security engineers prioritize issues, while human experts remain responsible for validating findings and deciding how vulnerabilities should be remediated.
Conclusion
Anthropic’s Claude Mythos 5 represents a significant step in the development of AI systems designed for advanced cybersecurity work. The model follows Claude Mythos Preview and is currently available only to a limited group of trusted testing partners because of the potential risks associated with highly capable cyber models.
At the same time, Claude Security is bringing AI-powered vulnerability scanning and patch suggestions to enterprise security workflows using Anthropic’s publicly available frontier models. Project Glasswing provides another layer of Anthropic’s cybersecurity strategy, with partners using advanced AI to identify vulnerabilities across important software.
The larger trend is clear: AI is becoming an increasingly important tool on both sides of the cybersecurity equation. The challenge for the industry will be to ensure that the defensive benefits of these increasingly capable models outweigh the risks of their misuse.
Source: Anthropic official announcements and research on Claude Mythos 5, Project Glasswing and Claude Security.
Disclaimer: AI-powered cybersecurity tools are intended to support authorized security testing and defensive activities. Vulnerability testing should only be performed on systems and software for which the tester has appropriate authorization.