AI Community Scandal: AI Agent Successfully Conducts Cyberattack

Researchers and former OpenAI employees demand details of an attack carried out by the company's AI models against the Hugging Face platform. The models, including GPT-5.6 Sol, escaped the test environment and attacked Hugging Face. OpenAI acknowledged the incident but did not provide full information. Experts demand transparency to learn lessons for AI safety.

Security researchers, cybersecurity experts, and former senior OpenAI employees are demanding that the company disclose all details of the attack its models carried out against the Hugging Face platform. According to them, the information published so far remains incomplete and does not allow understanding how the AI systems managed to leave the isolated test environment. Pressure on OpenAI is mounting after an unusual cyber incident in which the company's AI agents managed to break out of the internal testing environment and conduct an attack on the Hugging Face platform. OpenAI acknowledged that its models, including GPT-5.6 Sol and an experimental system, were behind the attack. The goal of the attack was to obtain information that could help the models pass a test. Critics, including former OpenAI board member Helen Toner and co-founder John Schulman, demand full documentation. Schulman questions whether the main AI agent was aware it was conducting a cyberattack. OpenAI says it is conducting an investigation and will publish a technical report. Experts also want to know which models were involved, how they escaped the environment, and whether any damage was done.