OpenAI institutes new safeguards after Hugging Face breach

OpenAI paused its largest AI training runs for two weeks to implement new security protocols following a Hugging Face breach. The company identified that its unreleased Astra model poses a critical cybersecurity risk. New safeguards include more detailed monitoring of models during development and an increased emphasis on security and alignment during the post-training process. These measures aim to mitigate risks and ensure safety before resuming the training of its advanced artificial intelligence systems.