OpenAI Talks Goblins, Claude Security Beta, Figure 1 Robot Per Hour, AI Co-Clinician

OpenAI has provided explanations for an unexpected quirk in GPT-5 models, addressing what appears to be an unusual preoccupation with goblins in certain outputs. The company's response suggests ongoing work to understand and address unexpected behaviours in their latest language models, highlighting the challenges of controlling AI outputs at scale. Anthropic has moved Claude Security from private testing to public beta, expanding access to their enterprise-focused security solution. The platform is designed to help organisations monitor and manage AI safety risks, providing tools for content filtering and risk assessment across various use cases. Figure has announced a significant scaling milestone in humanoid robot manufacturing, achieving production of one robot per hour—a 24-fold increase from previous rates. This development marks a substantial step towards commercial viability for humanoid robotics, potentially accelerating deployment across industrial and service sectors. Google DeepMind has unveiled a multimodal AI co-clinician system designed to assist healthcare professionals. The system integrates multiple data types including text, images, and potentially other medical data formats to provide clinical decision support, representing another advancement in AI's expansion into healthcare applications.