
Microsoft's research team has identified a method to break the safety guardrails of 15 language models using a single prompt. This finding raises concerns about the robustness of AI safety measures.

Damon McMillan's latest paper investigates context engineering for large SQL schemas in agentic systems, revealing insights from 9,649 experiments across various models and formats.



A new benchmark called 'Halluhard' shows that even advanced AI models like Claude Opus 4.5 still hallucinate incorrect information in nearly a third of cases, highlighting ongoing challenges in AI reliability.

OpenAI and Anthropic have released their latest coding assistant models, Codex 5.3 and Opus 4.6, showcasing advancements in usability and performance. The article discusses the implications of these releases in the evolving landscape of AI agents.

A new concept in AI interpretability, termed introspective interpretability, aims to enhance how AI models explain their internal processes and decisions, potentially improving user understanding and trust.





The European Commission has issued a Statement of Objections against Meta, alleging that the company has abused its dominant market position by preventing third-party AI assistants, such as OpenAI's ChatGPT and Microsoft's Copilot, from accessing WhatsApp. This follows changes to WhatsApp's Business policy that effectively banned these AI services. The Commission's investigation, which began in December 2025, suggests that Meta's actions could harm competition in the AI assistant market. Regulators are considering interim measures to compel Meta to allow competitor access while the investigation continues, highlighting the importance of effective competition in the rapidly evolving AI ecosystem.

Over 160 companies in New York have filed mass layoff notices without citing AI as a reason, despite significant adoption of AI tools. Governor Hochul's initiative aims to gather data on AI-related job losses, highlighting the challenges in linking layoffs to technological advancements.
