Can AI tell when someone is lying?

No — AI cannot reliably tell when someone is lying.

Some systems are marketed as lie detectors, but in practice they are not dependable enough to determine deception with confidence. What they usually do is look for patterns such as facial movements, voice changes, biometric signals, or language cues that may correlate with stress, arousal, or cognitive load — not lying specifically. An honest person can show the same signs if they are nervous, anxious, tired, or under pressure.

It is true that some commercial vendors claim high accuracy in controlled lab settings, but those claims do not mean the systems work reliably in real-world situations. Results can vary a lot depending on the person, the setting, the data, and how the system is used.

Because of these limits, AI lie-detection tools are generally treated cautiously, and they are often not considered reliable evidence on their own in legal settings. The EU AI Act also places strong restrictions on certain emotion-recognition and lie-detection uses, especially in sensitive contexts such as workplaces, education, and law enforcement.

So the short answer is: AI may sometimes flag behavior that looks suspicious, but it cannot reliably determine whether someone is lying.