AI coding tools have changed how quickly software can be written. A developer can now describe a feature in natural language, ask an AI model to generate the code and move from an idea to a working prototype far faster than before. But that speed creates a new problem for security teams. What if the code works — but contains a vulnerability that attackers can exploit? That is the problem Rubrik is trying to address with Rubrik Code Guardian, a cybersecurity capability designed to test code from an attacker's perspective. Rubrik first introduced Code Guardian in September 2026 in collaboration with Anthropic. The company has now expanded its Project Hourglass initiative around the technology, with additional global systems-integration partners joining the programme. The bigger idea is straightforward: if AI can help attackers discover vulnerabilities faster, defenders need security tools capable of analysing software at a similar pace..
What Is Rubrik Code Guardian?
Rubrik Code Guardian is designed to red-team source code using frontier AI. Instead of simply looking for individual coding mistakes, the system is designed to analyse relationships between different parts of a software environment. According to Rubrik, Code Guardian uses Anthropic's Claude Mythos 5 through a specialised Rubrik security harness. It works against a secure, air-gapped copy of a source-code repository rather than the live repository or production environment..
The tool is designed to identify:-
- Individual security vulnerabilities..
- Multi-step attack chains..
- Weaknesses spread across different services..
- Problems involving authentication and access controls..
- Risks crossing cloud and application boundaries..
- Vulnerabilities that may appear minor individually but become serious when combined..
That last point is particularly important..
Why Traditional Code Scanners May Miss Some Risks
Traditional application-security tools remain useful, but many are designed to identify individual weaknesses. For example, a scanner might detect an authentication problem in one component and a configuration weakness somewhere else. An attacker, however, may not look at those issues separately..
They may ask:-
Can these two weaknesses be combined?
A relatively low-severity vulnerability in one service could potentially become much more serious when combined with another weakness in authentication, permissions or cloud configuration. Rubrik says Code Guardian is designed to look for these multi-step attack paths, rather than treating every vulnerability as an isolated finding..
How Does Code Guardian Work?
The basic process can be understood in four stages..
1. Analyse the Codebase
The system examines source code across different files and components. Rather than limiting the analysis to one file, it can reason across services, authentication patterns and cloud boundaries.
2. Think Like an Attacker
The objective is not simply to ask whether a vulnerability exists. It is to determine whether different weaknesses can be connected into a realistic path that an attacker could use. This is what Rubrik describes as moving from individual alerts to attack chains..
3. Validate the Finding
A major part of the system is determining whether an identified attack path is actually exploitable. This is important because security teams already deal with large numbers of vulnerability alerts. A tool that produces thousands of additional theoretical warnings without establishing their practical importance could make the problem worse. Rubrik says Code Guardian is designed to validate attack chains before escalating them and rank findings according to exploitability, blast radius and business criticality..
4. Help Developers Fix the Problem
Once an issue is confirmed, Code Guardian can feed critical findings into existing developer workflows. Rubrik says confirmed critical issues can be pushed into systems such as Jira or GitHub, with file-level remediation guidance. The aim is to reduce the distance between: Finding a vulnerability → understanding its impact → fixing it..
Why Is AI-Generated Code a Security Concern?
AI-generated code is not automatically insecure. The problem is that AI can produce code at a speed that may exceed the speed at which humans can carefully review every line. Developers may also use AI-generated snippets from multiple sources or ask coding agents to modify several parts of an application. That creates several potential risks.
More Code, Faster
When software development accelerates, the volume of code requiring security review can also increase..
Human Review Can Become the Bottleneck
A developer may understand the feature they asked an AI tool to create but still miss a security implication elsewhere in the application..
Vulnerabilities Can Be Connected
An individual piece of code may appear harmless until it interacts with another service or permission boundary..
AI Agents Can Make Multiple Changes
As coding agents become capable of handling larger development tasks, security teams need visibility into not only individual code changes but also the wider architecture affected by those changes. This is why tools designed specifically around AI-era development are becoming an increasingly important part of application security..
What Is Project Hourglass?
Project Hourglass is Rubrik's broader initiative focused on the security and operational risks associated with autonomous AI agents. The company initially announced the initiative in June 2026 with global systems integrators including Cognizant, Deloitte, LTM, HCLTech, NTT DATA and Wipro.. On October 9, Rubrik announced an expansion of the initiative, adding AHEAD, Trace3 and WWT to the alliance. The company also announced the integration of Rubrik Code Guardian alongside Rubrik Agent Cloud. The broader objective is to help enterprises manage the security risks created as AI agents become more deeply involved in software development and business operations..
Why Does the Air-Gapped Environment Matter?
One of the most important aspects of Code Guardian is where the analysis happens. Rubrik says the system works on a secure, air-gapped copy of the codebase, rather than directly against a company's live repository or production environment. That matters because security testing itself can introduce risks if powerful tools are allowed to interact directly with production systems. An isolated environment allows security teams to investigate potentially dangerous attack paths without intentionally exposing live applications. It also means the same protected copy of the code can serve a second purpose. Instead of being used only for recovery after a cyber incident, Rubrik wants it to become an environment where security testing can take place before attackers find the weaknesses..
What Makes Code Guardian Different From a Basic Vulnerability Scanner?.
The key difference is the attack-path approach.
A conventional scanner may say:
There is a vulnerability here.
A more advanced system attempts to answer:
Can this vulnerability be combined with another weakness to create a realistic route into a sensitive system?
That distinction could become increasingly important as software becomes more complex and AI-generated code becomes more common..bRubrik says it has tested Code Guardian internally against its own codebase and found more findings than traditional code-scanning tools, while emphasising that the usefulness of those findings depends on validation and prioritisation..
Does This Mean AI-Generated Code Is Unsafe?
Not necessarily. AI-generated code should not automatically be treated as insecure simply because an AI model produced it. The bigger issue is how that code is developed, reviewed, tested and deployed. AI can also be used to improve cybersecurity. In fact, Code Guardian is an example of that approach: using advanced AI capabilities defensively to identify vulnerabilities before attackers can exploit them. The cybersecurity industry is increasingly moving towards this idea of AI versus AI — using AI's ability to analyse large amounts of software and identify complex patterns as part of defensive security..
What Should Developers and Companies Do?
A new security tool does not replace basic software-security practices..
Companies using AI coding tools should still maintain:-
- Human review of important code..
- Secure coding practices..
- Automated testing..
- Dependency and vulnerability monitoring..
- Access-control reviews..
- Secrets and credential protection..
- Continuous security testing..
- Clear approval processes before production deployment..
AI-generated code should be treated like any other code that enters a production environment: it needs testing and security review..
Is Rubrik Code Guardian Available to Everyone?
Not yet. Rubrik's official materials state that Code Guardian is currently in private preview and is accepting selected design partners. It is not yet generally available, and the company says features and availability may change. Rubrik previously indicated that general availability was targeted for fall 2026, but the current product information still describes the service as being in private preview. Therefore, it would be inaccurate to describe Code Guardian as a fully available consumer security tool..
The Bigger Shift in AI and Cybersecurity
The launch of Code Guardian reflects a larger change in the technology industry. AI is no longer being used only to generate text, images or simple pieces of code. AI systems are increasingly involved in software development, security testing, cloud management and other tasks that previously required significant human intervention. That creates a new cybersecurity race. Attackers can use AI to search for weaknesses faster. Defenders can use AI to find those weaknesses first. The advantage may increasingly go to organisations that can combine AI-powered analysis with strong human oversight, secure testing environments and reliable recovery systems...