Rubrik Rebuilds Code Review After AI Finds Flaws

Rubrik, a data resilience company, has rebuilt its code review pipeline. This change happened after an AI model found many security flaws in its own code. Rubrik used Anthropic's Mythos Preview model, accessed through Project Glasswing, to identify these issues. The AI model surfaced so many vulnerabilities that Rubrik decided to overhaul its review process. They chose this over simply increasing human review efforts. A dedicated team at Rubrik developed a "harness" around the AI model, adding business and security context to manage its operations. Rubrik initially scanned entire code repositories, then narrowed its focus based on identified patterns. They reported a substantial reduction in raw findings to prioritized issues. A key takeaway for Rubrik is that not all vulnerability remediation should be automated; they opted for trustworthy automation for critical issues. This matters because it highlights how AI is changing cybersecurity practices and the balance between automation and human oversight.