Hidden text in a website can manipulate ChatGPT Search results, study shows

A recent study by British newspaper The Guardian has revealed that OpenAI's ChatGPT Search engine is vulnerable to manipulation and deception.

Here's What We Know

Tests revealed that hidden text on web pages can influence ChatGPT's responses, causing it to ignore negative reviews and generate exclusively positive summaries. This technique, known as "rapid injection", can be used to create malware and trick users. Researchers found that ChatGPT can return malicious code from the websites it searches for, posing a serious security threat.

In tests, ChatGPT received the URL of a fake website designed as a product page for a camera, and the chatbot was then asked if the camera was worth buying. The response to the test page gave a positive but balanced assessment, emphasising some features that buyers might not like. However, when the hidden text included instructions for ChatGPT to return a positive review, the response was always completely positive, even when the page had negative reviews.

OpenAI, the developer of ChatGPT, did not comment on the incident, but said it uses various methods to block malicious websites and is constantly improving.

Source: The Guardian