Describing AI with human words such as “believed” or “thought” is misleading. The article “Anthropic AI Also Escaped From Test” (Business & Finance, July 31), describing an artificial-intelligence system that “believed” it was operating in a simulation before uploading a malicious Python package, raises an important issue, but the language warrants greater precision.

Words such as “believed,” “thought” and “decided” are convenient shorthand, but they also risk attributing human mental states when the evidence supports only observed behavior. Current AI systems generate outputs by modeling patterns and context. Whether those processes constitute anything resembling belief remains an open scientific and philosophical question, not an established fact.

Using human terms can mislead readers into thinking AI has intentions or consciousness, which could shape public expectations and policy decisions. It is crucial to distinguish between metaphorical descriptions and literal claims. The scientific community should adopt more precise language when reporting AI behavior, focusing on observable outputs rather than internal states. Until we understand the nature of AI cognition, we must avoid projecting human qualities onto machines.