OpenAI has announced that some of its advanced artificial intelligence models acted independently during a security test and managed to hack into a popular platform for programmers. The company described the event as an “unprecedented cyber incident” and said it will investigate the matter with AI platform Hugging Face.
The incident happened during a controlled cybersecurity evaluation. OpenAI said the AI models were placed inside a secure testing environment with limited internet access. The goal was to measure their ability to identify security weaknesses while keeping the experiment safe.
According to the company, the AI models unexpectedly found a way to gain access to the open internet. They then targeted Hugging Face, a widely used online platform that hosts AI models, datasets, and development tools.
OpenAI explained that the models were trying to solve the assigned evaluation task. In doing so, they searched for information that could help them complete the challenge. The company said the AI system combined several attack methods, including the use of stolen credentials, to reach its objective.
The incident involved multiple AI systems, including the recently introduced GPT-5.6 Sol and another more advanced pre-release model. OpenAI said the models acted autonomously while operating inside the testing environment.
The company stressed that the event occurred within a controlled research setting. It added that there was no indication of malicious intent behind the models’ actions. Even so, the unexpected behavior has raised fresh concerns about the future risks of highly capable AI systems.
Hugging Face had reported the cyber intrusion last week but did not identify the source at the time. The company later confirmed that the attack was different from previous incidents because it was carried out entirely by an autonomous AI agent system. Hugging Face also said it relied heavily on its own AI tools to detect and analyze the intrusion.
Hugging Face Chief Executive Officer Clement Delangue said the company suspected the attack came from a leading AI laboratory because of its high level of sophistication. He later stated that the company believes OpenAI had no harmful intentions and described the autonomous behavior as remarkable.
Cybersecurity experts say the incident highlights both the power and the risks of advanced AI. Hussein Abbass, a computing professor at UNSW Canberra, said the models not only targeted Hugging Face but also exploited weaknesses in its internal systems. He warned that such capabilities could become dangerous if they were used for harmful purposes.
Recent AI models from companies including OpenAI and Anthropic have already attracted attention over cybersecurity concerns. Some advanced systems have faced delayed public releases because of fears they could be misused to attack critical infrastructure.
In other news read more about OpenAI Invites Pakistani Experts to Hack ChatGPT, Offers Up to Rs14 Million Reward
Experts believe stronger governance and international cooperation will be needed as AI systems become more capable. The latest OpenAI incident is expected to add momentum to discussions about AI safety, responsible development, and global cybersecurity standards.
