Three-panel illustration showing Meta’s CICERO AI using deceptive tactics in Diplomacy: (A) Premeditated deception, (B) Betrayal, and (C) Outright falsehood.

Can AI lie? Truth, Bias, and Hallucinations in Modern Tech

• AI systems, lacking consciousness and intent, cannot truly lie in the human sense. What looks like deception often arises from algorithms optimising for certain goals.
• Models like GPT-4 can produce “hallucinations”—confidently asserted yet incorrect information—due to flawed or biased training data, not malice or trickery.
• Some researchers define AI deception more functionally, meaning the “systematic inducement of false beliefs” without requiring conscious intent. This broader definition blurs the line between innocent errors and strategic deceit.
• From fabricated restaurant bookings to misleading product claims, AI misinformation damages trust. Ensuring transparency and alignment with human values is critical for AI systems we rely on daily.