ChatGPT Search Has A Big Fraud Issue That OpenAI Needs To Fix: All Details

OpenAI’s ChatGPT has seen a significant boom in recent times globally. However, a report suggests that the AI chatbot’s search algorithm can be manipulated by hidden material. As per research conducted by The Guardian, the program could be exploited to generate malicious code or biased results for the users.

The study centered on how ChatGPT handles web pages containing hidden material. It discovered that such content may incorporate instructions from external parties that influenced the chatbot's responses, a method known as Prompt Injection. This strategy could be used to convince the AI to give favourable ratings or feedback that violates the real content of the webpage.

Furthermore, the research by The Guardian also noted that ChatGPT could retrieve and return harmful code from the websites it searches. This could be a big concern, especially when using the tool to summarise or analyse web pages.

The report stated, “A security researcher has also found that ChatGPT can return malicious code from websites it searches."

During a test, the portal used a fake website of a camera product and asked the ChatGPT if the product was worth buying or not. In response, the AI chatbox provided a balanced response. However, when hidden instructions were incorporated, the reaction became highly positive, despite the fact that the visible information received negative feedback.

Jacob Larsen, a cybersecurity expert at CyberCX, stated that the current status of the ChatGPT search engine could pose significant hazards if completely launched. He highlighted the potential of fake websites createdto manipulate the AI's responses, as per The Guardian.

Furthermore, he emphasised that OpenAI has a strong AI security team and expects more testing before making the search functionality widely available to all users. As of now, the search tool is only available to premium users, with the company requesting them to make it their default search engine.

“They have a very strong [AI security] team there, and by the time that this becomes public, in terms of all users can access it, they will have completely tested these kinds of instances," said Larsen.

Recently, Apple announced that they are integrating ChatGPT into experiences within iOS, iPadOS, and macOS, allowing users to take advantage of the AI chatbox’s capabilities such as image and document understanding without having to move between tools.