Anthropic's Mythos AI found a critical memory bug in OpenBSD's TCP code that had gone undetected for 27 years.
The episode begins with a satirical monologue about Americans living behind the moon and taking photos of a 'setting Earth,' framing the discussion with apocalyptic humor. Linus shares an anecdote about using the internet's correction mechanism to determine if Earth was waxing or waning in a photo, contrasting human and AI behavior. The conversation shifts to political commentary, briefly touching on Trump and Merz, before Linus recounts a childhood story about a broken 'Peugeot' sign and connects it to Dada artist Kurt Schwitters, who coined the term 'Merz.' The hosts then address serious policy issues, discussing proposed German legislation to reduce digital violence. They reference Anne Roth's presentation, which emphasizes the need for immediate victim relief over criminal penalties, and mention the 'Hexen' collective's anti-stalking resources. The episode critiques voluntary chat control measures and notes that companies like Google and Meta continue scanning despite legal changes. A significant portion is dedicated to a New York Times article using AI stylometric analysis to suggest Adam Beck is Satoshi Nakamoto, exploring the Cypherpunk movement and the implications of such identification. The climax focuses on Anthropic's 'Mythos' AI model, which demonstrated emergent abilities in security auditing, finding a 27-year-old memory corruption bug in OpenBSD's TCP implementation and generating exploits with 72% success rate. The hosts analyze the potential for AI to industrialize vulnerability discovery and exploit creation, predicting a paradigm shift in IT security, patch management, and software development practices. They conclude by reflecting on the need for new approaches to software maintenance and security in the face of rapidly advancing AI capabilities.
