AI technology faces recruitment manipulation and judgment limitations

Researchers at the University of Turin have studied a phenomenon known as prompt injection or keyword stuffing, where job candidates include invisible text in their CVs to manipulate AI recruitment tools. By using white text on a white background, candidates can insert keywords that are invisible to human recruiters but highly visible to algorithms. Tests conducted by the researchers showed this tactic achieved a 99% success rate with models such as ChatGPT, Gemini, and Claude.

Parallel to these security concerns, new research published on arXiv highlights a fundamental limitation in current large language models (LLMs): the lack of human-like judgment, intuition, and creativity. While AI can rapidly process scientific literature and execute experiments, studies suggest it fails when required to decide which hypotheses merit investment or when navigating open research that lacks predefined metrics. This distinction reinforces the necessity of human oversight in complex decision-making processes where objective verification is not immediately possible.