OpenAI has released a blueprint for the democratic governance of frontier AI, outlining a framework for how advanced artificial intelligence systems can be controlled and managed. This initiative aims to ensure safety and accountability through societal input and democratic principles.
This paper advocates for "open-world evaluations" as a complement to traditional benchmarks for measuring frontier AI capabilities. It introduces CRUX, a project for conducting these regular, long-horizon, real-world task assessments, exemplified by an AI agent successfully publishing an iOS app.
Anthropic announced Mythos, its most capable AI model, whose existence was revealed by a data leak from its servers. The model's release is highly restricted, and its disclosure negatively impacted cybersecurity stocks.
A significant tension has arisen between Anthropic and OpenAI regarding frontier AI security. Anthropic restricted Claude Mythos to a small corporate cohort, while OpenAI released GPT-5.5 to general availability, leading to comparisons of "Mythos-like hacking" open to all.
This paper addresses open problems in frontier AI risk management, highlighting the lack of scientific consensus and misalignment with existing practices. It systematically examines each stage of the risk management process, identifying challenges and classifying problems by their nature.
Google DeepMind is partnering with global consultancies. This collaboration aims to bring the power of frontier AI to organizations around the world.
