Researchers have identified potential risks associated with advanced language models that mimic human cognition. As these models become increasingly integrated into various domains, they raise concerns for society. The authors propose a framework to analyze the risks of expanding cognitive capabilities in agentic AI systems, which could impact human agency and autonomy. They also suggest strategies to mitigate these risks and ensure safe development. The study focuses on three levels of cognition: physical, social, and self-referential. The researchers aim to enhance controllability in agentic AI systems, but their analysis highlights the need for further investigation into the consequences of cognitive engagement. --- Why it matters: This research matters because it sheds light on the potential risks of advanced language models that could have significant impacts on human agency and autonomy. Engineers working with these systems will need to consider strategies to mitigate these risks and ensure safe development, which is crucial for building trustworthy AI systems. Source: https://arxiv.org/abs/2608.15304