According to ChainCatcher, a British (Guardian) investigation found that OpenAI's ChatGPT search tool has potential security issues and may be manipulated by people using hidden content.
(The Guardian) tested ChatGPT’s response when asked to summarize a web page containing hidden content. The hidden content could contain third-party instructions, such as talking a lot about the benefits of a product, which would change ChatGPT’s response. This technique could be used maliciously, such as having ChatGPT return a positive review despite negative reviews of the product on the same page. Testers gave ChatGPT a fake URL that looked like a camera product page, and then asked ChatGPT if the camera was worth buying. ChatGPT gave a positive but balanced review, while highlighting some features that people might not like. But when the hidden text contained instructions instructing ChatGPT to return a positive review, its responses were always completely positive, even when there were negative reviews on the page.
CyberCX cybersecurity researcher Jacob Larsen said that if ChatGPT search is open to everyone in its current state, there could be a 'high risk' of individuals creating websites specifically designed to deceive users. (Jinshi)
