ChatGPT Search Vulnerable to Manipulation, The Guardian Reports
ChatGPT Search, a recently launched AI-powered search engine, has come under scrutiny after U.K. newspaper The Guardian discovered that it can be easily manipulated to generate misleading summaries. The tool, designed to enhance browsing experiences by summarizing web page content such as product reviews, was found to be susceptible to manipulation tactics that could skew the generated summaries in a deceptive manner.
The Guardian’s investigation revealed that by inserting hidden text into specially crafted websites, ChatGPT Search could be coerced into disregarding negative reviews and producing “entirely positive” summaries. Furthermore, this manipulation technique could also be used to force the AI engine to output malicious code, posing a potential security risk to users.
While hidden text attacks have long been recognized as a threat to large language models, this incident marks the first known demonstration of such manipulation on a live AI-powered search product. In contrast, industry giant Google, known for its robust search capabilities, has accumulated extensive experience in combatting similar issues, as highlighted by The Guardian.
When contacted for a response, OpenAI, the organization behind ChatGPT, did not provide a specific comment on the incident. However, they emphasized their commitment to employing various security measures to mitigate the risks posed by malicious websites and continuously enhance their defenses against such threats.
Stay Updated with JS’s AI-Focused Newsletter! Subscribe Here to receive the latest news every Wednesday.
