Renato Vicente, Director at the TELUS Digital Research Hub, shares insights from a recent paper he co-authored on the security risks of persona-driven LLMs. The research finds that when LLMs are asked to adopt personas, a standard practice in most enterprise deployments, their moral judgments can shift significantly, with larger models proving more susceptible. Vicente says that organizations should put persona definitions through the same rigorous review as any other security policy and test for behavioral consistency, especially in high-stakes use cases like compliance, legal, or finance.