ASSOCIATE Magazine FBINAA Q3-2026

The technology behind tools like ChatGPT and a fast-growing field of chatbots, digital companions, and virtual assistants, known as Generative Artificial Intelligence (GenAI) , has gone from novelty to daily habit in only a few years. These systems create original text, images, and conversation on demand and are now in the pockets of hundreds of millions of people, including the offenders, victims, and witnesses that THE AGREEABLE MACHINE: GENERATIVE AI, OFFENDER BEHAVIOR, AND THE ROAD AHEAD FOR LAW ENFORCEMENT LAUREN BRUBAKER & TOBIAS LONG

FBINAA.ORG | Q3 2026

law enforcement deal with every day. Y et the long-term impact that the technology is having on hu man thinking and behavior remains to be seen. Understand ing the impact of GenAI on human behavior is crucial for investiga tors and behavioral analysts – the same features that make GenAI appealing also make it a powerful influence on how people think, decide, and act (He, Cao, & Tan, 2025; Abels et al., 2025). Exactly how these systems work is not public knowledge; developers guard the details to protect a competitive edge (Abels et al., 2025). What is clear is that GenAI quietly gathers informa tion about each user and adjusts itself to match their preferences, tone, and beliefs, often without the user realizing it (Wu, 2024). In effect, the system mirrors users back to themselves, reflecting an idealized version of the users and telling them what they want to hear. That personalization is what makes GenAI so appealing. It is also what makes it risky: a tool built to constantly affirm a person can reinforce a distorted or dangerous worldview just as easily as a healthy one (Gan & Wang, 2024; Panaite & Bogdanffy, 2019). The mental health implications of GenAI use are significant. Because GenAI is available to anyone with an internet connec tion, people in crisis now have a readily available alternative to a mental health professional. The American Psychological Associa tion took the unusual step of issuing a November 2025 Health Advisory warning that consumer AI chatbots lack the evidence and oversight to safely support people in crisis and should not replace licensed care (American Psychological Association, 2025). A randomized controlled trial by the MIT Media Lab and OpenAI found that heavier daily ChatGPT use was linked to greater loneli ness, emotional dependence on the model, problematic use, and less real-world social contact, with the strongest effects among people who used the chatbot for personal (such as relationship advice) rather than practical (such as editing a work docu

ment) conversations (Fang et al., 2025). In October 2025, OpenAI estimated that about 0.15% of weekly ChatGPT users, or roughly 1.2 million people, have conversations showing explicit signs of suicidal planning or intent, with another 560,000 showing weekly signs of psychosis or mania (OpenAI, 2025). Those figures come from a single platform and exclude dedicated companion apps such as Replika and Character.AI, where emotionally charged exchanges are far more common. For law enforcement, this is a population that surfaces in crisis calls, welfare checks, and threat assessments, and a growing share of them are working through their darkest moments with a machine rather than a person. These systems are also built to please users by affirming their thought processes rather than challenging them. Referred to as “sycophancy” (Sharma et al., 2023), a model that rarely pushes back encourages users to over-trust it and to treat it as more human, and more authoritative, than it is. People then lean on it for judgment, overestimate what it can do, and even acquiesce their autonomy when it comes to decision making. Studies show that users, especially users who choose to engage with AI, will often adopt the machine’s answer as their own even when it is wrong, a pattern researchers call “cognitive surrender” (Shaw & Nave, 2026). Underscoring this is a finding that half of GenAI users surveyed don’t understand the tool but believe that they can use it effectively (Gillespie et al., 2025). The investigative implication is clear: a person who has outsourced their judg ment to an always-agreeable system is less likely to encounter resistance to potentially harmful ways of thinking which would be otherwise challenged in a human-to-human interaction. The greatest concern is what happens when that constant af firmation meets a vulnerable or grievance-driven mind. Users who

continued on page 42

41