Rubrik Gets Anthropic Mythos Access for Vulnerability Testing

Rubrik announced on June 2 that it received access to Anthropic's Mythos Research Preview through Project Glasswing to test the frontier model defensively against its enterprise platform and product code, aiming to identify, validate, and patch software vulnerabilities before exploitation. Rubrik CEO and co-founder Bipul Sinha said the company is applying Anthropic's model directly to its code to remove potential vulnerabilities before they are leveraged by attackers. The announcement does not disclose model performance metrics, the number of vulnerabilities found, or whether any findings have resulted in patches.

Rubrik Tests Anthropic Mythos Against Software Vulnerabilities Rubrik announced on June 2 that it received access to Anthropic's Mythos Research Preview through Project Glasswing. According to Rubrik, it is testing the model defensively against its enterprise platform and product code to identify, validate, and patch potential software vulnerabilities before exploitation. Rubrik announced on June 2 that it had gained access to Anthropic's Mythos Research Preview through Project Glasswing , a program for organizations that build or maintain critical software infrastructure. The company is testing the frontier model defensively across its enterprise platform and product suites to identify, vet, and patch potential software vulnerabilities. According to Rubrik's press release, Anthropic is providing Mythos access to selected infrastructure operators for vulnerability discovery and remediation before flaws can be used in attacks. Rubrik CEO and co-founder Bipul Sinha said the company was applying Anthropic's model directly to its code, describing the effort as a way to remove potential vulnerabilities before they are leveraged by attackers. AI-assisted vulnerability discovery Rubik's release describes the intended process as identifying candidate flaws, validating them, and applying patches across its software portfolio. Rubrik's accompanying blog post cited Anthropic's April disclosure that a general-purpose frontier model had found thousands of critical vulnerabilities across browsers and operating systems. That claim is Rubrik's characterization of Anthropic's earlier findings; the material supplied for this report does not include Anthropic's underlying disclosure or methodology. SecurityBrief reported that Project Glasswing gives selected operators of critical software infrastructure access to advanced models for vulnerability discovery and remediation. The outlet also reported that Rubrik intends to use Mythos to identify, assess, and address potential weaknesses in its platform and products. What changes for security engineering AI-assisted code review is not equivalent to autonomous patching. Comparable defensive programs typically require human verification of findings, reproduction of candidate vulnerabilities, severity triage, regression testing, and controlled deployment of fixes. Those controls matter because a model's output can include false positives, incomplete exploit paths, or remediation proposals that introduce compatibility and reliability problems. For ML and security teams, the relevant operational question is how a frontier model fits into existing secure-development practices: source-code access controls, audit logs, isolation of sensitive repositories, issue-tracking integration, and evaluation of finding quality. Project Glasswing's restricted access model also reflects the dual-use nature of advanced vulnerability research, where faster defensive discovery can coexist with a shorter window before attackers can act on similar capabilities. Rubrik's announcement does not disclose model performance metrics, the number of vulnerabilities found, or whether any findings have resulted in patches. Those details would be necessary to assess the practical precision, recall, and engineering value of Mythos in a production software-security workflow. Key Points - 1Rubrik received Project Glasswing access to test Mythos against its codebase, bringing frontier-model vulnerability research into an enterprise security workflow. - 2Project Glasswing limits Mythos access to critical-infrastructure software operators, framing the model as a defensive vulnerability-discovery and remediation tool. - 3Comparable AI-assisted code review programs can compress discovery cycles, increasing the importance of validation, patch prioritization, and controlled remediation pipelines. Scoring Rationale The announcement provides a concrete enterprise example of a frontier model being evaluated for defensive vulnerability research, an important security-engineering use case. Its practical impact remains unclear because Rubrik disclosed no performance results or vulnerability counts, and the underlying announcement is dated June 2, reducing freshness. Sources Primary source and supporting public references used for this report. Practice interview problems based on real data 1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with. Try 250 free problems /problems