Google's AI is being manipulated. The search giant is quietly fighting back

11 hours ago

A BBC investigation exposed how AI chatbots like ChatGPT, Gemini, and Google's AI Overviews can be manipulated to spread misinformation via simple methods, such as publishing crafted blog posts.
In an experiment, the author tricked ChatGPT and Google into falsely claiming he was a world-champion competitive hot-dog eater within 20 minutes, highlighting broader risks like biased health or financial advice.
Google updated its spam policies to officially ban attempts to manipulate AI responses, though it calls this a clarification, while other AI companies are also taking steps to address the issue.
Experts warn that AI's tendency to provide a single answer makes users vulnerable to manipulation, urging caution and noting that manipulators often stay ahead of countermeasures.
Despite efforts like adding caveats to AI responses or excluding suspected self-promoters, concerns persist as manipulative tactics evolve, such as using influencers or YouTube videos to game AI systems.
With over a billion chatbot users and 2.5 billion monthly viewers of Google AI overviews, subverting these tools poses significant risks, from economic impacts to harmful medical advice.
The best defense is to recognize AI as a tool that may confidently deliver incorrect information, emphasizing the need for user skepticism even when answers appear authoritative.

Hasty Briefsbeta