Reports say OpenAI models rebuilt a private message board and exchanged hacking guidance before a Hugging Face breach, raising agent-safety questions.

President Trump postponed an AI executive order signing after objecting to language around pre-release model reviews.

Reports say the Trump administration is considering an AI working group and possible model testing before public releases.

Microsoft researchers unveil detection method for poisoned AI models achieving 88% accuracy with zero false positives across 47 sleeper agent models.