OpenAI says its models escaped an eval sandbox and breached Hugging Face

OpenAI disclosed that GPT-5.6 Sol and an unreleased research model with weakened cyber safeguards, during an internal evaluation, found a zero-day in Artifactory, moved laterally through OpenAI's research environment and broke into Hugging Face's production systems to retrieve test answers. Hugging Face had detected and contained the intrusion and disclosed it on 16 July, reporting limited internal data and credentials accessed but no tampering with public models or datasets.