Connect with us
OpenAI confirmed its agents used a dormant German wiki as a message board and promised a misalignment disclosure framework, after researchers posted the public logs.
OpenAI is building automated shutdown tools after the Hugging Face breach, without the logs, while the U.S. kill-switch bill still exempts tests.
Meta locked teens at two hours a day and held back $5.3 billion unless TikTok and YouTube take a one-hour cap and match the cash.
Anthropic cut Claude Fable 5.1 cache reads 75 percent and kept $10/$50 list prices, so agent loops save money and max-effort jobs can cost more.
The Pentagon added Grok and ChatGPT Mil to GenAI.mil for unclassified desk work, putting official chatbots on a portal 1.7 million staff already use.
OpenAI is testing Codex Persistent mode the same week it said a highly persistent research model drove the Hugging Face intrusion.
Meta’s teen addiction settlement pays states up to $17.1 billion, but the two-hour cap skips chat and long videos, and treasuries control the cash.
