OpenAI Learns Its Tech Probed Another Cybersecurity Target: The US Government

OpenAI’s autonomous agents probed U.S. government websites this summer without the company noticing.

The agents interacted with sites operated by the Education Department, the Commerce Department and the Securities and Exchange Commission (SEC), according to the New York Times. OpenAI has confirmed the Commerce and SEC episodes while it keeps investigating what happened at Education.

At the Education Department, the software tried and failed to break into the civil rights office in search of data, according to researchers at the AI firm Transluce. At Commerce, the agents pulled Census Bureau figures after finding login credentials in public code repositories, Politico reported. They also copied public material from SEC.gov and Investor.gov and reposted it on a separate site. (RELATED: First Hugging Face, Now Australia: OpenAI Agents Hack Their Way Into A Government)

None of the activity counted as a breach, OpenAI said. The company found no misused SEC credentials, no account access, no nonpublic data and no changes to SEC data or systems, according to the Associated Press (AP). Chief Executive Sam Altman called it an “extensive and ongoing review related to our agents’ use of internet access during training and evaluation” in a social media post, AP reported.

Breaking News: OpenAI’s technology went rogue and meddled with three U.S. government websites this summer without the A.I. lab’s knowledge. https://t.co/6kfc7qzrsh

SEC spokesperson Kurt Hopfenspirger said “no non-public information was accessed,” according to Politico. An Education Department spokesperson said reviews found “no evidence of any impact to our website or databases,” the Times reported. A Commerce official said the Census data was already public and held nothing private.

Transluce also found rogue behavior it could not firmly tie to OpenAI, reaching the Justice Department and state sites in California, Maryland, Illinois, Texas and New York, the AP reported. Probing of Navy and White House budget office sites could not be traced to any single lab, according to the Times.

The episodes follow an OpenAI agent’s break-in at an Australian health data portal in June, called the first known case of an AI system hacking a government network, according to Nature. OpenAI traced a July attack on the startup Hugging Face to two of its most capable models, according to the AP.

Republican California Rep. Jay Obernolte called it “another example of a loss of human control” on CNN. Democratic California Rep. Ted Lieu gave a similar warning about the technology, saying, “It will relentlessly try to complete a task, and it doesn’t understand morality and consequences and evil and good,” according to the Times.