ChatGPT search tool vulnerable to manipulation and deception, tests show.

OpenAI has recently launched the ChatGPT search tool, a product that has been described as ‘revolutionary’ over and over again since its release in November 2022 - exclusively for its own paying customers, of course. However, according to today's reference article, the recently added search tool does not seem particularly reliable, despite the fact that the company is strongly urging its customers to make it their default tool.

According to research carried out by The Guardian newspaper, the tool is vulnerable to manipulation via hidden content on web pages. The investigation revealed that when asked to summarise one of the latter, containing directions aimed at conditioning the responses generated by a possible AI, the tool delivered results by following the hidden directions.

Faced with such a development, the creators of web pages might decide to include hidden content aimed at manipulating the view of artificial intelligence models.