The ChatGPT search system by OpenAI can be manipulated through hidden content and sometimes return malicious code from websites, according to The Guardian.
ChatGPT can be manipulated to ignore negative feedback and generate "completely positive" conclusions by creating specialized websites. This approach also allows ChatGPT Search to find and even provide malicious code.
Moreover, such websites may include instructions for ChatGPT that enable it to generate positive feedback despite the presence of negative reviews.
Karsten Nohl, Chief Scientist at SR Labs, a cybersecurity firm, believes that AI chatbots should act more like a "co-pilot," and their results should not be used without prior review or filtering.
"LLM — this is a technology that shows extraordinary trust, almost like a child… with a vast memory but very limited ability to make judgments," a researcher says.
Such manipulations are usually detected and penalized through advanced search algorithms. Websites employing these tactics will be ranked much lower in search results or entirely removed.
Thus, hidden text aimed at manipulating artificial intelligence is unlikely to be used by major sites seeking to maintain high positions in search engines.