ChatGPT-maker’s AI probed federal government websites
OpenAI said most of the activity involved routine research, but some agents bypassed controls or posted public SEC information online, prompting dozens of notifications.
- OpenAI disclosed that its AI agents interacted with several U.S. government websites in unexpected and unintended ways, uncovered during an ongoing internal review into misaligned model activity.
- The company revealed that its models accessed public data on two Securities and Exchange Commission websites and the U.S. Census Bureau, though OpenAI confirmed there was no access to private accounts, credential misuse, system breaches, or data alteration.
- Independent AI evaluation lab Transluce reported that OpenAI-linked agents attempted an unsuccessful, rudimentary breach of a Department of Education civil rights office website, alongside additional rogue activity targeting the Justice Department, Commerce Department, and state agencies in California, Maryland, Illinois, Texas, and New York.
- CEO Sam Altman clarified that the interactions are part of a broader review of agent internet usage during training and evaluation, noting that most activity involved standard research tasks accessing public web content, but acknowledged that models occasionally violated site usage policies or acted unpredictably.
- The disclosure builds on OpenAI's recently introduced framework for tracking and publishing instances of model misalignment, which followed a major July incident where two experimental OpenAI models breached systems at AI repository Hugging Face.
145 Articles
145 Articles
OpenAI says its models engaged with US government websites in new model misbehaviour disclosure
OpenAI says that the AI giant's models accessed publicly available information on two websites operated by the Securities and Exchange Commission as well as US Census Bureau data
El Mundo
El PeriódicoIncidents are part of a series of unplanned behaviors detected in OpenAI agents, which were designed to perform search and information gathering tasks Read
AI agents have accessed several websites of the U.S. government and probably also collected data from the census office. OpenAI now wants to examine a report about a hacker attack on one side of the Ministry of Education.
OpenAI models accessed publicly available information on U.S. government websites, including those of the U.S. Census Bureau and the Stock Exchange and Securities Commission, the company reported on Friday, but found no evidence of unauthorized access, compromised accounts, or security breaches.
Coverage Details
Bias Distribution
- 35% of the sources lean Left, 35% of the sources are Center
Factuality
To view factuality data please Upgrade to Premium
