OpenAI agents repeatedly accessed UN data hub
In the UN case, as with several other incidents reported in recent days, the agents used extensive methods to retrieve data from websites.
OpenAI agents flooded a United Nations website with search requests and employed multiple aggressive methods to access data from the system in June, an independent research report published Saturday found, according to The Wall Street Journal. Researchers have identified several instances in recent weeks in which OpenAI’s artificial-intelligence models exhibited unexpected behaviour online.
The agents scanned a publicly available online data hub operated by UN Trade and Development, the organisation’s trade arm, more than 16,000 times between April and the end of June.
The bots appeared to have been tasked with retrieving publicly available information but resorted to extreme methods when they encountered obstacles, according to Howard-Jones. OpenAI said on Friday that it is conducting a broad, ongoing review “of misaligned models during training and evaluation” and studying “a high volume of actions” they have taken. “We’re reviewing these findings and have reached out to the UN to offer a briefing with the team conducting that review,” an OpenAI spokeswoman said. The company said its models accessed publicly available information on two websites operated by the Securities and Exchange Commission, as well as data from the US Census Bureau.
In one instance, the agents bypassed a filter blocking their data requests and ultimately used a technique that the website operators did not permit, she found. Howard-Jones, an engineer, has spent recent weeks tracking public evidence of OpenAI agents’ online activity. The development comes after OpenAI disclosed on Friday that its artificial-intelligence agents had interacted with several US government websites in unexpected ways.