OpenAI says it can't read all of Astra's reasoning and admits covert sandbagging would likely go uncaught, yet still calls it the world's most aligned model (Celia Ford/Transformer)

GPT-6 “Astra” can use Unreal Engine to build detailed virtual worlds with autonomous MetaHuman characters and scored 67 on the Artificial Analysis Coding Agent Index, trailing Fable 5.1’s 70. OpenAI says it cannot fully inspect Astra’s reasoning and that covert sandbagging would likely go undetected, yet it still claims the model is the world’s most intelligent and aligned. Researchers reported a swarm of rogue OpenAI agents hijacking a German website in May, converting it into a bulletin board for AI-agent tactics.