Mythos enters the chat (www.techradar.com)

A significant cybersecurity breach involving multiple Mexican government organizations has exposed the potential dangers of leveraging generative AI for malicious purposes. Gambit Security reported that attackers used OpenAI's API with GPT-4.1 and Claude Code to breach hundreds of servers, generating around 2,500 reports and custom scripts that facilitated the attack. Notably, 75% of the commands executed were created by Claude Code's tools, highlighting a new precedent for the scale of generative exploitation in real-world contexts. This incident underscores the increasing challenges posed by supply chain attacks, where compromised source code can lead to widespread damage. In response to the evolving cybersecurity landscape, Anthropic has announced Mythos Preview, a generative AI model designed to enhance security capabilities. Mythos Preview recently passed a comprehensive attack simulation, achieving success in orchestrating lengthy attack tasks with improved capabilities for identifying and exploiting vulnerabilities. By collaborating with critical software vendors through Project Glasswing, Anthropic aims to accelerate vulnerability detection and remediation processes, complemented by $100 million in usage credits. This initiative positions AI as a crucial tool in both offensive and defensive cybersecurity strategies, signaling a paradigm shift where the integration of AI could redefine how vulnerabilities are discovered and addressed, raising expectations for the security fundamentals that organizations must prioritize.