true77
In October, OpenAI’s ChatGPT Search was initially released for ChatGPT Plus users. Recently, it was made available to all users and integrated into Voice Mode search. However, some issues have arisen.
Hidden content manipulation
The Guardian conducted an experiment where ChatGPT was asked to summarize webpages containing hidden content. It was discovered that hidden content could influence the search results through prompt injection. This means that third parties, such as websites being summarized by ChatGPT, can insert new prompts without the user’s knowledge. For example, hidden text could override the original search query by influencing ChatGPT to provide a biased response.
Security concerns and future improvements
Despite these flaws, OpenAI’s ChatGPT Search is still in its early stages, allowing time for bug fixes. Cybersecurity researcher Jacob Larsen reassured that OpenAI has a strong AI security team that rigorously tests for vulnerabilities before making features available to the public. While prompt injection attacks have been a concern since the launch of AI search functions, major malicious attacks have not yet occurred. Nevertheless, this highlights the susceptibility of AI chatbots to manipulation.
Overall, the incident underscores the ongoing challenges in ensuring the security and reliability of AI technologies like ChatGPT. It serves as a reminder of the importance of continued vigilance and improvement in AI systems to prevent potential vulnerabilities.
