Deezer: Over 50% of Daily Uploads Now AI-Generated

OpenAI has published findings on safety and alignment challenges specific to long-horizon AI models, documenting new risks, observed failures, and improved safeguards developed through iterative deployment. The company shares lessons learned from operating these extended-capability systems in production environments. The work addresses practical safety concerns that emerge when models operate over longer time horizons and decision chains.

OpenAI has introduced GPT-Red, an automated red teaming system that uses self-play to identify and address vulnerabilities in AI models. The system is designed to improve safety, alignment, and robustness against prompt injection attacks. GPT-Red represents an approach to proactive AI security testing that could inform how organizations evaluate model vulnerabilities before deployment.

OpenAI has outlined a 'reverse federalism' approach to AI governance in which state-level laws work together to establish a national framework for safe and democratic AI development. The approach positions state action as foundational to building coherent federal policy rather than the traditional top-down model. This represents a shift in how the AI industry and policymakers are thinking about regulatory coordination across jurisdictions.

DeepMind CEO Demis Hassabis is calling for the creation of an independent standards body to regulate frontier AI models, modeled after FINRA, the financial industry regulator. The proposed body would test frontier models before release and develop best practices for their deployment. Hassabis's proposal addresses growing concerns about how advanced AI systems are evaluated and released to the market.