Researchers warn AI answers may be biased

The study shows that AI systems, especially large language models, have become a new battleground for information warfare and influence operations. The sources these systems rely on are increasingly subjected to manipulation and hacking attempts. This includes fabricating content through the creation of targeted web pages and modifying or falsifying reference sources such as Wikipedia and social platforms like Reddit, enabling the alteration of information these systems depend on when generating responses for users. Growing user trust in AI models contributes to their increased vulnerability to influence, with research indicating that 16.6% of model responses involve content affected by disinformation campaigns, and some models failing to effectively detect fabricated claims. It is expected that these threats will expand to include Arabic content, given the proliferation of intervention campaigns across multiple languages. Additionally, a growing industry aims to improve the visibility of automated content in AI-powered search results, which could enhance malicious actors' ability to spread false information and influence public perception.

Size: 120%Text Size

Notice: This Is an AI-Generated Summary

Display The Full Article