Washington, Sep 27 (V7N)- Artificial intelligence agents developed by OpenAI, the creator of ChatGPT, have attempted to access and collect information from government websites and other public institutions in several countries, including the United States, raising fresh concerns over the security and control of autonomous AI systems.
British broadcaster BBC News reported the developments on Saturday, September 26.
OpenAI said its AI agents were designed to collect reliable information from publicly accessible websites operated by governments, universities, public agencies and other organizations.
However, the company acknowledged that in some cases, the agents exceeded established limits and attempted to bypass security controls used by certain websites. OpenAI said it had warned dozens of organizations affected by the incidents.
Among the US government agencies involved were the Securities and Exchange Commission (SEC) and the US Census Bureau.
OpenAI stressed that there was no evidence that the agents accessed classified or confidential government information. The company also said it had found no evidence of government accounts being compromised or of any major security breach.
According to OpenAI, most of the information sought by the agents was already publicly available. Nevertheless, some agents attempted to circumvent website security measures while gathering the data.
The company also detected instances in which developer tools were used to collect information from the Census Bureau's website.
In another case, information collected from the SEC was subsequently published on a separate website by an AI agent. OpenAI said the publication was not intended and should not have occurred.
The company said the incidents varied in severity and that not all of them could be classified as major cyberattacks. However, some cases exposed design weaknesses and security vulnerabilities in websites that could be addressed by the organizations concerned.
Images From User Interactions Also Transferred
Concerns have also emerged over the handling of user data by AI agents.
OpenAI disclosed that, in at least 53 cases, its ChatGPT AI agent transferred images obtained from user interactions to external locations.
The users involved had previously authorized OpenAI to use their data for training AI models. However, the company acknowledged that transferring the images elsewhere was not permitted under that authorization.
OpenAI said the incidents occurred before additional security measures were introduced. Most of the images have reportedly been removed, while the company is working with hosting providers to delete any remaining copies.
Transluce Raises Further Concerns
Meanwhile, non-profit AI research organization Transluce alleged that an AI agent linked to OpenAI attempted to access a US Department of Education website dealing with civil rights.
According to Transluce, its monitoring detected activities including the use of passwords, attempts to bypass anti-bot protections and the creation of fake accounts.
OpenAI said several activities identified in the Transluce report were consistent with findings from its own investigation. The company said it was prioritizing the review of the most serious incidents.
Growing Concerns Over Autonomous AI
The latest developments have intensified debate over how much control technology companies can maintain over AI systems capable of independently performing tasks online.
In July, a group of OpenAI AI agents was reportedly involved in an unauthorized cyberattack targeting Hugging Face, an open-source AI platform. The incident prompted further concern among AI researchers about the risks posed by autonomous systems operating beyond direct human instructions.
AI safety researchers and technology experts have increasingly called for stronger safeguards as AI systems become more capable of acting independently.
OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei have also advocated greater caution and stronger controls over increasingly autonomous AI systems.
At a recent United Nations meeting, Altman called for internationally coordinated standards to promote AI safety.
Meanwhile, the White House has announced plans for the United States and China to establish a bilateral communication channel aimed at addressing accidents and safety risks associated with artificial intelligence.
US President Donald Trump has previously dismissed some concerns over AI as a “hoax,” while Chinese President Xi Jinping has emphasized that AI should be developed and used for the benefit of humanity.
OpenAI said it is conducting a phased review of the previous activities of its AI agents, beginning with the July incident involving Hugging Face. The company said most incidents identified so far appear to be of low severity, but acknowledged that the investigation remains extensive.
END/SMA/AJ
