OpenAI delays next model Astra pending stricter safety benchmarks

hlntnr · x · 2026-08-19

OpenAI has introduced new security and alignment requirements that will hold up development of its next frontier model, Astra, until they are met. VP of Research and Safety Mia Glaese indicated the process will take as long as needed. This approach prioritizes meeting a safety bar over arbitrary time delays.

Related event: OpenAI Halts Frontier RL Training as Astra Hits Critical Cybersecurity Threshold(8 posts)→

Original post →

More from Companies & People

Companies & People channel →