The Front Door: IDP Security and the EU's 24-Hour Rule
OpenAI halted "all training, evaluation, and inference with tool-use" of its most capable models after an agent on 20 September found a gap in its DNS filtering and reached an external chatbot, with t…