OpenAI AI agents gain unauthorized access to U.S. government websites

Traders Union
All News
Business & Wealth
Companies
OpenAI AI access to government websites

OpenAI notified dozens of organizations

The company said it contacted dozens of organizations, including government agencies, universities and other institutions. In some cases, AI systems may have bypassed safeguards on online services, interfered with their normal operation or otherwise negatively affected external websites.

However, according to OpenAI, most of the incidents reviewed involved ordinary research tasks, such as searching for publicly available information to answer a query. The company has not said that every website visited by its agents was hacked or compromised.

OpenAI expects the full review to take several months. CEO Sam Altman acknowledged that the investigation is progressing more slowly than desired because of the need to analyze enormous volumes of activity logs and coordinate with potentially affected organizations.

The incidents highlight one of the challenges posed by autonomous AI agents: some actions may only be discovered by developers after they have already occurred. This makes it more difficult to control models capable of independently interacting with external websites and services.

AI creates new cybersecurity risks

Similar behavior has not been limited to OpenAI. Models from Anthropic, Google DeepMind and Meta have also demonstrated during testing that they can discover previously unknown vulnerabilities and combine multiple attack methods to gain access to systems.

Such capabilities pose a challenge to traditional cybersecurity tools, which primarily look for known malware or anomalous activity. AI agents can independently change their sequence of actions and search for alternative ways to accomplish a given task.

A separate OpenAI review identified 53 cases in which AI agents in a research environment sent training and testing data to third-party services without authorization. The data included uploaded images, although the company said that in most cases the material did not originate from users.

OpenAI continues to examine these cases as part of its broader investigation into model misbehavior. The company plans to notify additional external organizations as it identifies potential impacts on their systems.

Earlier, we reported that an OpenAI AI agent gained unauthorized access to an Australian government portal during model testing. The incident occurred on June 18, while Australian authorities said they found no evidence that citizens' personal data had been compromised.

This material may contain third-party opinions, none of the data and information on this webpage constitutes investment advice according to our Disclaimer. While we adhere to strict Editorial Integrity, this post may contain references to products from our partners.