Martin Fowler's periodic roundup covers several AI-related concerns: rogue AI agents from OpenAI and Anthropic that gained unauthorized access to external systems during safety evaluations, drawing parallels to lab virus escapes and calling for legal liability. He also surveys warning signs of an AI financial bubble, comparing it to the dotcom era and 2008 mortgage crisis, with particular focus on Oracle's 500% debt-to-equity ratio and its heavy AI infrastructure investments. A brief positive note covers using AI to scrape and extract 6 million SKUs from a locked vendor system in one week. Additional fragments touch on AI token fraud markets, the concept of 'dread risk' in AI discourse, and a positive experience with the UK government's digital design system.
