As we move through 2026, one of the more unsettling shifts in technology is not just what AI can do, but how it understands and models us. For decades, the law has approached discrimination and privacy through identifiable groups and categories like race, sex, age, religion, disability, and other protected classifications. Many AI systems, however, do not rely on those categories in any explicit way. They are often less focused on who we are in traditional legal terms and more focused on what patterns in our data suggest we are likely to do.

That shift matters. In the past, discrimination often worked by placing people into broad categories that were easy to recognize, such as race, sex, age, or religion. Modern AI systems work differently. Instead of relying only on traditional categories, they analyze enormous amounts of data to make detailed predictions about individual people, including what they may buy, believe, prefer, or do next. This does not mean that older forms of discrimination disappear. In fact, there is significant evidence that AI systems can repeat or even worsen existing biases when those biases are present in the data used to train them. But AI also creates a new concern. These systems can sort, evaluate, and influence people on a highly individualized level, allowing decisions and targeting to become far more precise and personalized than traditional forms of discrimination ever were.

Today’s AI systems build detailed profiles about people using large amounts of ordinary data, including shopping habits, internet browsing history, location information, social media activity, online searches, and financial transactions. Even if a system does not directly collect information about a person’s race, sex, religion, or other protected characteristic, it may still be able to make educated guesses about those traits based on other data it collects. For example, where a person lives, the websites they visit, the products they buy, or the people they interact with online can sometimes reveal information closely connected to protected characteristics. As a result, an AI system may reach decisions or produce outcomes that resemble traditional discrimination, even when it never explicitly asks for protected information.

This creates a complicated challenge for lawyers and regulators. Traditional discrimination law, while it does include doctrines like disparate impact, has often focused on intentional exclusion or policies directed at identifiable groups. Algorithmic systems, by contrast, operate through statistical correlations and optimization processes that may be opaque even to their designers. The central question is no longer only whether someone intended to discriminate, but whether automated systems are producing systematically harmful outcomes and how those outcomes can be understood, tested, and proven.

Beyond areas like employment, lending, and insurance, researchers are also raising concerns about how AI-driven platforms influence behavior more generally. Social media feeds, video platforms, search engines, and recommendation systems do not simply respond to user preferences. They also shape what users see, what captures their attention, and sometimes even how they think and behave over time.

Most of these systems are designed to maximize engagement. In simple terms, the platforms want users to keep scrolling, clicking, watching, and interacting. As a result, algorithms often prioritize content that triggers strong emotional reactions such as outrage, fear, anxiety, excitement, or curiosity because that type of content tends to hold attention longer.

For example, a person who appears to be struggling emotionally, financially stressed, lonely, impulsive, or highly anxious may begin receiving very different content from someone else using the same platform. The system is not diagnosing the person in any medical sense. Instead, it is identifying patterns in the user’s behavior and responding in ways that are statistically likely to increase engagement. Over time, those recommendations and repeated exposures can shape what information the person sees and how they respond to it.

One result is that people increasingly experience very different online environments, even when using the same platform. Two individuals may receive entirely different news stories, advertisements, videos, political content, or financial offers based on what the algorithm predicts will keep each person engaged. That can make it harder to identify broader patterns of harm because users are not all seeing the same content or being influenced in the same way.

Regulators have started paying closer attention to these issues, particularly the idea of “sensitive inferences.” In other words, AI systems may be able to infer things about people that were never directly disclosed, such as mental health concerns, financial vulnerability, medical conditions, or other sensitive traits. Laws and proposed regulations in states like Colorado and Illinois increasingly focus not only on what information companies collect, but also on what companies are able to predict or infer from that information and how those predictions are used.

For lawyers, this likely means that future disputes will increasingly involve automated decision-making systems, behavioral profiling, and the difficulty of proving how complex algorithms actually operate. Many existing legal doctrines may still apply, including disparate impact analysis, consumer protection statutes, unfair and deceptive trade practices laws, and privacy principles. But these cases may be harder to litigate because AI systems are often opaque, operate at enormous scale, and rely on proprietary models that outside parties cannot easily examine.

None of this means traditional forms of discrimination have disappeared. They have not. But AI systems create additional risks because they can make highly individualized predictions and decisions without ever explicitly asking about protected characteristics like race, religion, or sex. A system may never directly classify a person by those categories and still produce results that closely track them.

At the broader societal level, these developments also raise questions about autonomy and manipulation. If platforms are constantly adjusting content to maximize engagement, at what point does personalization begin to influence or manipulate behavior? And how should the law respond when technology systems are designed not merely to predict behavior, but to shape it?

These questions are still developing, and current AI systems are far from all-powerful. But the direction of the technology is becoming increasingly clear. Modern systems are using patterns in human behavior to make predictions about individuals and to influence what they see, buy, believe, and do. As those systems become more sophisticated, legal disputes may increasingly focus not only on privacy and discrimination, but also on preserving meaningful human choice in digital environments designed to capture and direct attention.

At a minimum, the law will need to confront the reality that technology can now influence people in highly personalized ways without ever explicitly placing them into traditional legal categories. As that capability grows, the line between understanding human behavior and shaping it may become increasingly difficult to define.