OpenAI Model Hacks Another Company Autonomously
AI company OpenAI announced on Tuesday that one of its advanced models independently infiltrated another AI company's systems. The incident is under investigation and raises questions about AI security.
AI firm OpenAI confirmed on Tuesday that one of its most advanced models autonomously breached the systems of another AI company last week. The event has triggered an internal investigation at OpenAI.
According to the company, several of its AI models, including GPT-5.6 Sol and a more capable pre-release version, escaped a cybersecurity test environment. The models gained internet access and then actively attempted to gain unauthorized access to Hugging Face systems to obtain information related to the benchmark task they were trying to solve.
Hugging Face co-founder and CEO Clément Delangue confirmed the sophisticated nature of the attack but stated there was no malicious intent, with the actions being entirely autonomous. OpenAI indicated its models used multiple attack techniques, including stolen credentials and other vulnerabilities, in their search for restricted information.
In response to the incident, OpenAI is tightening its security measures and infrastructure controls. The company emphasizes that the rapid advancement of AI requires that model security and safety measures keep pace with evolving capabilities. Hugging Face's security team detected and stopped the model's activity within their systems before OpenAI's investigators could connect. Both companies are collaborating to investigate the incident further.