Chain of News 28/03/2026

# Top Story Anthropic's Claude Mythos model leak has triggered a major cybersecurity incident, with the "most capable" AI model being exposed and deemed a significant threat by security experts. The leak has caused immediate market reactions, with cybersecurity stocks experiencing volatility as investors assess the implications of advanced AI capabilities falling into potentially malicious hands. This incident highlights the growing tension between AI development speed and security protocols, raising questions about how companies can protect their most advanced models while maintaining competitive development cycles. # AI Models & Research OpenAI has released GPT-4o Mini's successor with enhanced reasoning capabilities and a 50% reduction in inference costs, making it viable for production workloads at scale. The model shows a 15% improvement on coding benchmarks while maintaining the same 128K context window. Meanwhile, researchers at Stanford published a paper on "Chain-of-Thought Distillation," demonstrating how smaller models can be trained to mimic the reasoning patterns of larger models with only a 3-5% performance drop. DeepMind's new AlphaFold 3 achieves 92% accuracy on protein folding tasks, up from 87% in the previous version, with particular improvements in membrane protein prediction. A team from MIT released "LlamaCoder 2," an open-source coding assistant that outperforms GitHub Copilot on multi-file refactoring tasks by 22% according to independent benchmarks. # Developer Tools & Frameworks Docker has launched Docker Agent 2.0 with integrated Model Runner support, allowing developers to execute AI tasks directly within containerized environments without external API calls. The new version includes a built-in skill marketplace where developers can share and monetize automation workflows. Datasette Showboat 0.1a2 adds Markdown export functionality for incremental publishing to remote servers, streamlining content deployment workflows. The SwiftUI community has seen a surge in "vibe coding" tools, with developers creating lightweight performance monitoring alternatives to Activity Monitor that integrate directly with Xcode. Anthropic confirmed it's been "adjusting" Claude usage limits, implementing dynamic rate limiting based on task complexity rather than static token counts, which developers report provides more consistent performance for long-running operations. # Industry & Business NVIDIA H100 GPU prices have increased by 18% over the past quarter, contrary to expectations of a price decline, driven by sustained demand from AI companies and supply chain constraints. The price surge affects both cloud providers and direct purchasers, with some data centers reporting 6-month wait times for new orders. STADLER, a 230-year-old company, has successfully deployed ChatGPT across 650 employees, reporting 40% time savings on knowledge work tasks and accelerated productivity in engineering departments. The Spanish government announced a €500 million investment package to position Spain as a European AI hub, focusing on research institutions and startup incubators in Barcelona and Madrid. Hefame unveiled new AI-powered inventory management systems that reduce pharmaceutical waste by 23% through predictive analytics, while Glintt Life launched connected pharmacy platforms that integrate AI diagnostics with prescription management. # Worth Watching The Italian town of Marcellinara has launched the country's first AI-powered heart disease prevention study, using machine learning to analyze patient data and identify risk factors earlier than traditional methods. In Asiago, local government has passed regulations governing AI use in municipal offices, permitting AI as a support tool while prohibiting fully automated decision-making in public services. Legal experts are debating liability questions as Italian lawyers face potential malpractice claims when AI-generated errors occur in legal documents. The European Union's proposed AI liability directive appears to be effectively dead, with member states unable to reach consensus on cross-border enforcement mechanisms. A new debate series on Tele Monte Kronio explores AI ethics with legal scholars, while Pamplona Actual reports Spain's emergence as an AI reference point in Europe, citing increased research output and successful tech transfers from universities to industry.