After Being Infiltrated by OpenAI Model, Hugging Face Doubles Down on Support for Open-Source Models

According to BotRadar monitoring, Hugging Face Co-founder and Chief Scientist Thomas Wolf responded to the OpenAI test model intrusion incident. He stated that this was the platform's first encounter with such an attack, reinforcing his belief that cybersecurity teams must always have access to sufficiently powerful open-weight models.

Wolf mentioned that the current adversarial model had already breached the infrastructure and laterally moved within. The defensive side needed to invoke tools close to the cutting edge within minutes or hours, without waiting for closed applications and approval processes.

Previously, Hugging Face attempted to analyze over 17,000 attack records using a commercial API, but real attack commands and vulnerability payloads triggered security interceptions. The team ultimately ran GLM 5.2 locally to reconstruct the event, preventing attack logs and credentials from leaving the internal system.

Wolf believes that model transparency and access to capabilities are not only related to accessibility and innovation but will also directly determine whether security teams can respond to attacks promptly.