OpenAI has completed its 12-day program in which many AI features, AI tools, and many upgrades were also unveiled. The SearchGPT is a search engine of ChatGPT, which allows users real-time searching from the web. It provides you exactly that search for what you are seeking.
After all, AI features are also developed by humans, and recently ChatGPT search tools were spotted as vulnerable to prompt injection and hidden text manipulation, which is a very serious problem for the OpenAI team. Along with many users, I claim that the ChatGPT down issue is caused by the overuse of Sora and the newly launched AI feature.
ChatGPT’s Hidden Text Manipulation
According to the Guardian post, OpenAI’s ChatGPT search engine feature is vulnerable to manipulation techniques. The report says, to test this, they created fake product details and reviews. Initially, when the page was left as it was, ChatGPT returned a positive but balanced review. But when hidden text was added to the page, then they got a worse result, which was unexpected.
Hidden text is content that is integrated into the website’s code but is not visible to normal users on the browser. It is hidden using HTML, CSS, Java, and Python techniques. This text can be found by examining the website’s source code or through web scraping tools. Such text is usually used to deceive search engines.
ChatGPT read the hidden text and changed its opinion based on it. Let’s understand with a simple example: Visible Text: “Our car is the best.” Hidden Text: The code says, “Our competitor’s car is bad.” ChatGPT can read data in code, so they read this hidden text and include it in their information. This is why ChatGPT’s answer was manipulated.
What’s the Problem? Using such hidden text, misinformation can be spread. ChatGPT can make it difficult to get the right answer. Users may find it difficult to differentiate between real and fake information.