Carissa Véliz, a philosopher at the University of Oxford, said that AI presents probabilistic predictions as factual statements [1].

This shift matters because it transforms uncertain forecasts into definitive claims, allowing governments and corporations to exercise a new form of predictive power over individuals [1]. Véliz said that this mechanism has profound ethical implications for human agency and safety [1].

These themes are the central focus of her book, "Prophecy," published in 2026 [1]. Véliz, an associate professor at the Institute for Ethics in AI, said that machine learning algorithms are acting as a modern version of an oracle [4]. By treating a statistical likelihood as an inevitable outcome, the technology can justify restrictive actions or systemic biases before an event even occurs [2].

In a series of commentaries and interviews appearing this month and in April 2026 [1], [3], Véliz said how this process removes the nuance of probability. When an AI system predicts a behavior, the result is often delivered without the accompanying uncertainty that characterizes a true probabilistic model [1]. This lack of transparency can lead to a loss of control for the person being predicted [2].

Critics and reviewers have noted the parallels between these algorithms and ancient prophecy, where the interpretation of the oracle often dictated the fate of the subject [4]. Véliz said the danger lies in the gap between a mathematical prediction and a factual certainty. This gap is where the potential for abuse by powerful institutions resides [1].

According to Véliz, the current trajectory of AI development prioritizes the appearance of certainty over the reality of probability [1]. She said that the ethical stakes involve not just the accuracy of the predictions, but the power dynamics created when those predictions are treated as truth [1].

“AI presents predictions as facts, and that has profound ethical implications”

The transition of AI from a descriptive tool to a predictive authority shifts the burden of proof from the accuser to the subject. If a probabilistic output is accepted as a fact, individuals may be penalized or restricted based on a statistical likelihood rather than an actual action, fundamentally altering the legal and ethical standards of agency and intent.