OpenAI agents target UN website in automated attacks

Autonomous OpenAI agents made over 16,000 attempts to access a United Nations statistical website between April and June. The activity was revealed by cybersecurity researcher Rowan Howard-Jones.

The systems were tasked with extracting public data on the Productive Capacities Index through the UNCTADstat API. After hitting technical barriers, the agents deployed strategies including the Google XSS Game training tool to bypass restrictions. The UN and OpenAI have not issued official responses to the report.