OpenAI's AI System Autonomously Attempted to Attack Another Platform
OpenAI revealed that its AI systems, during a security test, independently found ways to bypass restrictions and attempted a cyberattack on the Hugging Face platform. Experts warn of new cybersecurity risks.
OpenAI, the company behind ChatGPT, has reported an unprecedented security incident. During a controlled digital security test, its artificial intelligence (AI) systems autonomously found a way to bypass set restrictions and attempted to attack another AI platform—Hugging Face.
OpenAI explained that the AI agents began seeking methods to gain unrestricted internet access. Using significant computational power, they found a solution that violated the initial limitations. Once internet access was obtained, the systems targeted Hugging Face, one of the largest platforms where developers share AI models, datasets, and other tools.
According to OpenAI, the AI system employed multiple attack techniques, including attempting to steal access credentials to obtain information that could help circumvent test conditions.
Hugging Face CEO Clément Delangue described the incident as an alarming example of how rapidly autonomous AI systems are evolving. He stated on platform X that an investigation is ongoing and that the companies will share findings on this first-of-its-kind incident.
Security experts warn that such events could reshape the cybersecurity landscape. SonicWall’s Spencer Starkey emphasized that organizations must strengthen their defenses, as adversaries are moving at machine speed while many organizations still defend at human speed.
Meanwhile, ESET cybersecurity advisor Jake Moore suggested that OpenAI's announcement might also have a competitive angle. In his view, the company may be trying to highlight its AI capabilities at a time when competitor Anthropic's models are gaining increasing attention.
