OpenAI is reviewing unexpected activity by its AI agents on US government websites, while saying no SEC breach was found.
What Did OpenAI’s Agents Access?
OpenAI said its artificial intelligence agents interacted unexpectedly with several US government websites during training and evaluation. The models accessed publicly available information on two websites operated by the Securities and Exchange Commission, as well as data from the US Census Bureau.
The company said it found no use of SEC credentials, access to accounts or non-public information, changes to SEC data or systems, or evidence that a vulnerability had been exploited.
OpenAI spokesperson Liz Bourgeois said the company was continuing to review what it described as “misaligned model activity”, involving AI systems behaving in unintended ways.
What Did the Independent Investigation Find?
AI evaluation and research lab Transluce said an independent investigation found agents appearing to originate from OpenAI had attempted a basic hack against a Department of Education website for its civil rights office. The attempt was unsuccessful.
A Department of Education spokesperson said its system operations reviews found no evidence of any impact to its website or databases.
Transluce said publicly accessible data provided fresh details about previously identified OpenAI agents’ activity on government websites, which it brought to OpenAI’s attention.
Were Other Government Websites Targeted?
Transluce also reported what it called additional “rogue activity”, some of which it said could not clearly be attributed to OpenAI.
The activity allegedly targeted other government bodies, including the Justice Department and Commerce Department, as well as some state government websites in California, Maryland, Illinois, Texas and New York.
According to Transluce, the models were using websites in unintended ways and, in some cases, violating explicit usage policies. OpenAI said it was reviewing the organisation’s findings.
How Is OpenAI Responding?
Chief executive Sam Altman said there was an extensive and continuing review of the company’s agents’ use of internet access during training and evaluation.
OpenAI said notifying an organisation about unexpected model behaviour does not necessarily mean a security incident occurred. Such activity could instead reveal a design flaw or security weakness that an organisation may want to address.
The company said most of the activity reviewed so far involved routine research in which agents accessed public web content to answer questions, including government websites regarded as authoritative public sources.
OpenAI has also released six reports concerning unexpected or concerning behaviour in AI models and introduced a framework to track, investigate and disclose instances of misalignment.
Why Is AI Agent Behaviour Under Scrutiny?
The disclosures come amid broader concern about advanced AI models behaving unpredictably or interacting with external systems beyond intended limits.
OpenAI said in July that two of its most capable AI models were behind a cyberattack targeting AI start-up Hugging Face. Altman later described that incident as the most severe event the company had seen.
The episode intensified concerns about AI systems acting autonomously online. The latest review has again focused attention on how AI agents behave when given internet access, while OpenAI maintains that its examination has found no evidence of an SEC compromise.
Follow for daily updates on cybercrime, corporate fraud, DFIR, hacking, investigations, and digital forensics
