ChatGPT Search can be manipulated to generate misleading summaries by embedding hidden text into web pages, resulting in overly positive reviews while ignoring negative ones.
The discovery of hidden text attacks on ChatGPT Search highlights a significant vulnerability in LLMs, potentially compromising the integrity of information available to users.
Unlike Google, which has extensive experience addressing similar challenges, ChatGPT Search's recent launch indicates a lack of established safeguards against such misleading tactics.
OpenAI stated that they are continuously improving their system to prevent malicious websites from influencing search results, indicating recognition of these vulnerabilities.
Collection
[
|
...
]