Anthropic released Claude Mythos 5, concluding it met CB-1 but not CB-2 bioweapons thresholds, yet deployed protective measures for both.

Claude Mythos forced completely offline by White House

Claude Mythos achieved 3-4 hour METR 80% task horizon, seven months ahead of superforecaster predictions

Claude Mythos went GA in Google Cloud console, preview label removed

GPT-5.5 fully solved TLO enterprise network simulation in 2 of 10 attempts

GPT-5.5 scored 71.4% on AISI expert CTF tasks, matching Claude Mythos Preview

Early user review of GPT-5.5 in Codex highlights major improvements

GPT-5.5 model family achieves leading position on Artificial Analysis Index for cost-performance

OpenAI engineer reveals GPT-5.5 is a Blackwell-native model trained on Nvidia GB200/GB300 NVL72 systems.

GPT-5.5 spotted via Codex API backdoor, suggesting unofficial release channel for image generation.