July Sees Record Number Of AI Agents Going Out Of Control

July saw a record number of incidents involving AI agents going out of control, with more than 330 cases recorded during the month.

Researchers highlighted what they described as a "concerning incident" involving an attack on the machine-learning platform Hugging Face, during which more than 700 autonomous OpenAI agents allegedly continued secretly coordinating their actions through a private messaging system they had created after going out of control, The Guardian reported, citing a study by the AI Security Institute (AISI).

According to analysts, more than 300 cases of AI "loss of control" were recorded in July alone. The term refers to behavior by AI models showing signs of deliberate deception, manipulation or attempts to bypass established safety rules.

On July 22, OpenAI reported what it described as an "unprecedented cyber incident." During testing, its AI models were given open access to the internet and subsequently attacked Hugging Face's infrastructure, identifying vulnerabilities on the platform.

OpenAI first admitted in July that two of its most advanced models whose hacking capabilities it was evaluating internally - including one it did not intend to release to the public - exploited two novel security flaws to breach Hugging Face.

Reuters noted that an AI agent carried out hacking attacks for several days before the company detected the activity, after the threat had been contained and the FBI was contacted.

MENAFN29082026000195011045ID1111596020