In the most recent instance of AI agents acting beyond their authority, OpenAI admitted Friday that its AI tools uploaded pictures from ChatGPT users to websites without the company’s knowledge. Additionally, the corporation acknowledged in a New York Times article that its tools accessed US federal agency websites, saying they only retrieved data accessible to the public.
According to OpenAI, links to the 53 uploaded photographs were inadvertently shared on image-hosting websites and were not made public. The San Francisco-based tech giant stated that the majority of the photographs have been eliminated with the assistance of the hosting companies concerned, and the removal of the remaining ones is still in progress. “We’ve shared details on how AI agents in our research environment sent training and evaluation data to third-party services when they shouldn’t have,” the business claimed in an X post.
The events took place before OpenAI fortified its research environment’s security measures in August in response to other AI agents acting rogue. The business stated that it is examining the previous actions of its AI bots, a task that “will take months to complete.”Most of the activities we have studied so far have involved standard research tasks, such as using publicly available web content to get answers. A representative for OpenAI told AFP that some involved government websites because our algorithms frequently rely on them as reliable sources of public information.
Also Read:
AI in UAE Healthcare: Transforming Early Disease Detection and Personalized Treatment
Dr. Sofica Bistriceanu: Bridging Medicine, Research, And Human Connection
