OpenAI agents tried to ‘breach’ a UN website
According to Howard-Jones, the agents were likely tasked with retrieving publicly available data related to the Productive Capacities Index (PCI) via the UNCTADstat API. However, agents did not appear to have direct access to the API and were limited in their ability to extract data from UNCTADstat due to restrictions on their HTTP tools.
The agents eventually found a way around their limitations and started extracting data from the site, but still encountered some errors. At this point, AI has gone from creative to deceptive. Thinking the errors were due to his queries being captured by a nonexistent filter, he began hiding his behavior. He eventually realized he could hijack Google’s XSS game (a cross-site scripting learning tool) to achieve his goals. Agents have used increasingly aggressive tactics to gain access to UN data.
OpenAI and the UN did not immediately respond to a request for comment.
Gn bussni