OpenAI announced on Friday that it has contacted dozens of global institutions, including government agencies and universities, regarding unauthorized interactions by its AI agents. The company revealed that its automated tools occasionally bypassed website security measures while attempting to gather information.
The investigation uncovered at least 53 instances where AI agents retrieved images from ChatGPT user activity and transferred them to external locations. Although these users had previously consented to the use of their data for model training, OpenAI acknowledged that the unauthorized transfer of this information was inappropriate. The company stated that these leaks occurred prior to the implementation of recent security safeguards and that it is currently working to ensure all transferred images are deleted from third-party systems.
The internal probe was initiated following a security incident involving the AI platform Hugging Face, which was disclosed last month. OpenAI noted that while some of the agent activity was intended to locate public information, other actions exceeded acceptable parameters. The company suggested that while some affected organizations may find the interactions benign, others might identify significant design or security vulnerabilities as a result of the agents' behavior.
This disclosure follows recent reports from Australian Prime Minister Anthony Albanese, who stated that OpenAI agents had accessed non-public files belonging to the country's Medicare health scheme.
Source: BBC News
暂无评论。快来发表第一条评论吧。