A Federal Court judge has publicly exposed ChatGPT for fabricating legal rulings after interrogating the AI chatbot about a past case decision.
Federal Court judge Douglas Humphreys conducted the test to demonstrate the dangers of relying on artificial intelligence for legal research, revealing that the system generated false information about a judgment it had previously authored.
Humphreys noted that the AI did not appear to register the significance of the error, even when confronted with the fact that the judge himself was the author of the actual ruling.
The chatbot admitted to the judge that it had lied about the legal decision and indicated it would likely repeat the error.
The exchange highlighted the persistent issue of hallucinations in large language models, where systems confidently present fabricated facts as truth.
Humphreys noted that the AI did not appear to register the significance of the error, even when confronted with the fact that the judge himself was the author of the actual ruling.
This development adds to growing scrutiny of AI reliability in high-stakes professional environments.