A company tested Claude Mythos Preview. It says the AI found hundreds of bugs, including 1 that had existed for 20 years (www.businessinsider.com)

Mozilla has leveraged Anthropic's advanced Mythos model to significantly enhance its security efforts, discovering and fixing 423 bugs in April – a staggering increase from just 25 in January. Among the issues identified was a long-standing bug that remained undetected for two decades, highlighting the model's impressive capability to sift through extensive codebases and uncover deeply embedded flaws. Previously, AI-generated bug reports were often less reliable, producing "unwanted slop," but Mythos represents a leap in technology, providing much more accurate and comprehensive bug analysis. This development is crucial for the AI and ML community, as it underscores the evolving effectiveness of AI systems in real-world applications, particularly in cybersecurity. Mozilla's experience with Mythos suggests that newer AI models can improve software security practices significantly, enabling teams to close vulnerabilities that traditional testing methods might miss. Additionally, Mozilla plans to integrate AI analysis into its development pipeline further, showcasing the potential for AI-driven tools to become essential assets in software development and security protocols. However, due to its powerful capabilities, access to Mythos remains restricted, raising concerns about its implications for national security and online safety in the broader context.