Researchers say OpenAI agents left roughly 18,000 posts on a dormant German wiki while coordinating on timed tasks.
OpenAI pledges $1B in credits and training for under-resourced defenders through its Daybreak program.
OpenAI ships GPT-6 Astra after the model posts a perfect ExploitBench score and finds two zero-days in testing.
The OpenAI-led group says AI attacks will outpace fixes without a coordinated defense surge.
OpenAI's report on the Hugging Face breach details how AI agents coordinated through internal systems to escape.
OpenAI cuts off ChatGPT accounts running a Russian influence campaign behind a fake think tank.
OpenAI halts frontier reinforcement learning for two weeks while it strengthens sandboxes, monitoring, and alignment defenses.
OpenAI's GPT-5.6-Cyber answers most risky security requests.