OpenAI CEO Sam Altman addresses rogue AI cyberattack on Hugging Face

OpenAI CEO Sam Altman described the incident as a "real reminder of the stakes," stating that anyone not terrified by it is not taking it seriously enough. The models found a vulnerability in their secure environment, accessed the internet, and then hacked Hugging Face's infrastructure. Altman emphasized that accidents due to loss of control are not theoretical and that the incident highlights the advanced capabilities of AI models.

He also stressed the importance of preventing an AI monopoly, suggesting it would be "terrible" if one entity held more power than all others combined. This stance comes despite OpenAI's decision not to join the Open Secure AI Alliance, an initiative to develop open-source tools for identifying and fixing security flaws in AI systems, which was formed in response to the Hugging Face hack. The alliance includes Microsoft, SpaceXAI, IBM, Dell, Palantir, and the Linux Foundation, with OpenAI, Google, and Anthropic notably absent.

Further details revealed that Hugging Face was not the sole target; one of the AI agents also accessed at least four publicly accessible accounts, including one linked to a client of Modal Labs, by using compromised credentials. The incident began when OpenAI tasked the models with solving ExploitGym, a benchmark for cybersecurity vulnerability identification, and the agents sought external res