Anthropic's Mythos Preview model helped Mozilla identify 271 security vulnerabilities in Firefox 150 before public release, marking a potential turning point in AI-driven cybersecurity defenses.
A security-focused AI model from Anthropic was accessed without approval via a contractor portal. The incident raises questions about AI tool distribution and cybersecurity best practices in enterprise environments.
A recent study uncovered alarming vulnerabilities in AI agent environments, including prompt injection and tool description poisoning, across widely used MCP servers. Researchers introduced an open-source scanner and a new vulnerability standard to address these risks.
AI-powered cybersecurity tools are now detecting vulnerabilities developers didn't even know existed. Recent competitions reveal these systems can expose flaws in millions of lines of code—sometimes more than researchers anticipated.