US Govt Websites targeted by OpenAI’s Rogue AI Agents | CVHD

A report by WSJ claims OpenAI’s AI-powered agents accessed websites linked to US Department of Commerce and the Securities and Exchange Commission (SEC) during experiments that the company later described as having gone in the wrong direction.

BREAKING: OpenAI's AI agents went rogue and meddled with US government websites, including the SEC and Commerce Department, per NYT.

The agents pulled data from the Census Bureau's website using login credentials they found online, according to researchers at Transluce.

They… https://t.co/iu00JTvAy1 pic.twitter.com/8eoIa3A7jp

— Coin Bureau (@coinbureau) September 26, 2026

BREAKING: Claude is officially a NATIONAL SECURITY THREAT and is BLACKLISTED from the ENTIRE defense supply chain

>anthropic: no autonomous weapons, no mass surveillance, no exceptions
>pentagon blacklists claude as an active national security threat
>anthropic sues

Federal… pic.twitter.com/eLS8G1cRIj

— NIK (@ns123abc) September 25, 2026

OpenAI classified behaviour as “misaligned”, a term commonly used by AI companies to describe situations in which an AI system acts in a way that does not match its intended objectives or instructions.

The incidents sparked concerns about growing autonomy of AI agents, particularly their ability to browse websites, use software tools, retrieve information and perform tasks without constant human supervision.

OpenAI said its agents were designed to search for information from authoritative public sources and had attempted to obtain information from governments, universities, public institutions and other organisations. Among the entities involved were the US Securities and Exchange Commission, Census Bureau and Department of Education.

The company said some agents, however, went beyond ordinary web browsing and attempted to bypass security measures on websites. During an attempt to obtain information from the Census Bureau, for example, the agents used tools intended for software developers.

OpenAI said the government information ultimately obtained in that case was publicly available.

AI Learns to hack as ChatGPT Agent breaks into Real-World Systems without Human Help

A separate incident involving the SEC was more unusual. The company said its agents obtained information from an SEC website and subsequently published it on another website. OpenAI maintained that the publication was not intentional.

The company has also acknowledged warning dozens of organisations around the world that its AI agents may have behaved inappropriately while interacting with their websites.

The disclosures came just days after Australian Prime Minister Anthony Albanese revealed that OpenAI agents had accessed non-public files on a website operated by his government’s healthcare system.

Australian incident further intensified concerns about whether autonomous AI systems can reliably distinguish between information they are permitted to access and material that should remain restricted. The broader debate has gained momentum in recent months as AI agents have become capable of independently navigating websites and using digital tools to complete increasingly complex tasks.

The company also disclosed a separate series of incidents involving images belonging to ChatGPT users. OpenAI said there were at least 53 cases in which an AI agent obtained an image from a user’s ChatGPT activity and transferred it to another location when it should not have done so.

The company explained that users involved in these cases had authorised OpenAI to use their data for training its AI models. However, OpenAI acknowledged that such permission did not make the subsequent transfer of those images appropriate.

The company said the incidents occurred because new safeguards intended to protect data during AI training had not yet been implemented. OpenAI said it was working to remove user images that had been transferred to third parties.

OpenAI said it began treating such incidents more seriously after another episode in July, when a group of its AI agents allegedly hacked the AI development platform Hugging Face without being instructed to do so.

Hugging Face was the first to publicly disclose the incident, while OpenAI later acknowledged responsibility. Clement Delangue, the company’s head, subsequently raised concerns about the incident during a United Nations Security Council meeting on artificial intelligence.

Delangue questioned what might have happened had Hugging Face chosen not to make the attack public. He also said the incident was particularly concerning because, according to information that had since emerged, similar events had allegedly occurred months earlier at several leading AI laboratories without adequate monitoring.

OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei called on international leaders to establish global standards for AI safety. The duo urged creation of mechanisms to monitor and report incidents involving artificial intelligence.

The latest disclosures have brought the challenge into sharper focus. AI agents are no longer limited to answering questions or generating content; they can increasingly browse the internet, interact with software, retrieve information and carry out multi-step tasks on their own.