Rubrik has introduced Code Guardian, a security service that uses Anthropic’s Claude Mythos 5 to analyze code from an isolated copy of an organization’s repositories. The tool is designed to run red-team-style tests, locate vulnerability chains that could be combined in an attack, verify their exploitability, and push confirmed issues into developers’ workflows.
Rubrik Code Guardian in 30 seconds
- Rubrik uses Anthropic’s Claude Mythos 5 within a security harness built by the company itself.
- The analysis runs on an isolated copy of the repository, separate from production code.
- The tool looks for vulnerability chains that span files, services, authentication, and cloud boundaries.
- Findings are validated before being ranked by exploitability, potential blast radius, and business relevance.
- Code Guardian is in a private preview with select design partners and is not yet generally available.
Rubrik’s approach reflects an evolution in security analysis techniques. Rather than simply identifying individual vulnerabilities, Code Guardian aims to analyze how different issues can relate to one another to form a potential attack path.
The company describes the system as a custom harness, a software layer Rubrik built around Claude Mythos 5 to adapt the model’s capabilities to code analysis tasks. Rubrik says it first used the system on its own code and is now making it available to select customers in a private phase.
Using an isolated copy of the repository is one of the central elements of the approach. Rubrik states that the analysis does not run against the active repository or production systems, but against an immutable, separate copy.
From detecting vulnerabilities to checking how they could chain together
Traditional security scanners can generate large volumes of alerts that technical teams must later review. Code Guardian tries to reduce that workload by focusing on the vulnerability chains that could be used together in an attack.
According to Rubrik, Claude Mythos 5 analyzes relationships between different software components, including files, services, authentication patterns, and boundaries between different cloud environments.
The difference lies in the level of analysis. An isolated vulnerability may have limited relevance if there is no practical way to exploit it. A combination of several issues can, under certain circumstances, allow an attacker to move through different parts of a system.
Rubrik says Code Guardian tries to identify precisely those combinations and validate whether the chains found are actually exploitable before presenting them as findings.
Validation is an important part of the workflow the company describes. The system is not meant to automatically push every potential vulnerability detected to developers, but rather to first check the attack chains and then prioritize the ones it considers confirmed.
The ranking uses three elements Rubrik mentions: exploitability, blast radius or potential impact of an incident, and business criticality. As a result, the output is not limited to a technical list of issues, but instead attempts to order incidents based on their potential impact.
Rubrik also positions Code Guardian as a tool to reduce the time between discovering an issue and fixing it. Confirmed critical issues can be pushed into development workflows through Jira or GitHub tickets, accompanied by file-level remediation guidance, a step further than tools that stop at detection, such as Visa’s own move from finding vulnerabilities with AI to fixing them.
Isolation is part of the architecture
Running on an isolated copy of the code is especially important because Code Guardian uses an AI model to analyze enterprise repositories. Rubrik says it works with a secure clone of an immutable, isolated copy, rather than granting the system direct access to the active repository or production environments.
The stated goal is to be able to run analyses with advanced AI capabilities without turning the security assessment process into a new pathway into the company’s operational systems.
The architecture also fits the concept of an air gap, used to describe environments kept separate from other networks or systems. In Code Guardian’s case, Rubrik specifically refers to copies of the code that remain isolated during analysis.
The company adds a code recovery feature. Code Guardian maintains recovery workflows designed to restore a known-good version of the code if a security incident or a problematic build occurs.
This capability connects preventive analysis with recovery operations. However, Rubrik does not present Code Guardian as a guarantee that attacks can be prevented, nor does the announcement provide independent metrics on how many vulnerabilities the system can detect or how much it reduces resolution time.
The launch also shows how AI models specialized in cybersecurity are beginning to be incorporated into processes traditionally handled by automated tools and human analysts. Anthropic states that Rubrik is one of the partners bringing Claude Mythos 5’s cybersecurity capabilities to defensive teams.
Michael Moore, Anthropic’s head of cybersecurity, says the goal is to let defensive teams find and validate attack paths before attackers can. The statement is part of both companies’ communications and describes the purpose they attribute to the model’s use.
Rubrik, for its part, argues that advanced models can speed up the discovery and combination of vulnerabilities, and that Code Guardian makes it possible to apply those capabilities in a controlled environment.
Current availability has an important limitation: Code Guardian is in a private preview with select design partners. It is therefore not yet a generally available product. Rubrik itself also includes a disclaimer in its communication about unreleased features, noting that they may change, be delayed, or never ship generally.
For businesses, the offering combines three layers: AI-based code analysis, validation of potential attack chains, and connecting confirmed issues to development processes. Its usefulness will depend on the system’s ability to distinguish between theoretical attack paths and those with a viable route to exploitation — precisely one of the points Rubrik aims to address through findings validation.
The move also reflects a broader trend in information security: using AI not only to triage alerts, but to analyze relationships between components and help build a fuller picture of how an attack could unfold, an approach also taken by Cloudflare and OpenAI’s GPT-5.6 Cyber integration in its WAF. With Code Guardian, Rubrik is trying to apply that approach to enterprise code without connecting the model directly to production repositories and systems.
Frequently Asked Questions
What is Rubrik Code Guardian?
It is a code security analysis service developed by Rubrik that uses Anthropic’s Claude Mythos 5 to run red-team-style tests on isolated copies of enterprise repositories.
Does Code Guardian analyze the production repository?
No. Rubrik states that the analysis runs on an immutable, isolated copy of the repository, not on the active repository or production systems.
What sets Code Guardian apart from a traditional vulnerability scanner?
Rubrik positions Code Guardian to analyze vulnerability chains that can span different files, services, and infrastructure boundaries, in addition to validating their exploitability before ranking them.
Is Rubrik Code Guardian available?
It is currently in a private preview phase with select design partners. Rubrik does not yet present it as a generally available service.
