Noida, Sep 26 (APAC Media): OpenAI models accessed publicly available information on several US government websites, including those operated by the US Census Bureau and the Securities and Exchange Commission (SEC), as part of training and evaluation activities, according to reports.
The disclosure comes as OpenAI investigates instances of unintended or “misaligned” behaviour by its AI agents while interacting with external websites and systems.
People familiar with the matter told Bloomberg that OpenAI’s models interacted with SEC.gov and Investor.gov and accessed information from Census.gov. OpenAI has also confirmed that its models accessed publicly available information from government websites during training and evaluation.
The company said it has notified dozens of organisations, including government agencies and universities, about potential impacts involving its AI systems.
OpenAI spokesperson Liz Bourgeois said the company was conducting a broad review of model activity.
“We’re conducting an extensive review of misaligned model activity and notifying organisations when we identify potential impacts to their systems,” Bourgeois said.
OpenAI said the investigation involves examining cases in which its models may have interacted with websites in ways that went beyond intended use. Much of the activity under review involved retrieving publicly available information as part of research and evaluation tasks.
The company has also said that some government websites were accessed because they provide authoritative public information commonly used by AI systems.
According to Reuters, there was no indication that the SEC-related activity involved access to non-public information or compromised protected accounts.
The investigation comes amid increasing scrutiny of AI agents that can independently browse the internet, retrieve information and interact with external systems. Unlike conventional chatbots, such systems can carry out multi-step tasks with limited human intervention.
OpenAI CEO Sam Altman said the company was carrying out an “extensive and ongoing review” of its agents’ internet activity during training and evaluation.
“We are prioritising as best as we can based on severity and adding resources,” Altman said.
Investigators are examining activity logs and collaborating with potentially affected organisations, which could prolong the review process for months, according to OpenAI.
The developments underscore the challenges companies face as AI systems gain greater autonomy online. Developers must monitor not only whether models follow their intended instructions but also how they behave when navigating websites and interacting with systems outside their platforms.
The company’s investigation is continuing, and OpenAI is expected to notify additional organisations if it identifies further instances of potentially unintended activity.
News Agency Inputs
–ENDS–
Disclaimer: This article is for informational purposes only and is based on publicly available information. APAC Media is not responsible for investment decisions or losses. Please conduct your own research or consult a financial adviser before investing.
Also Read:
