A.I. Hallucinations Are Rising As Tools Grow More Complex

New A.I. systems from companies like OpenAI are more advanced than ever, but they’re increasingly spreading false information — and nobody knows why.

A.I. systems are becoming more powerful, but they’re also making more mistakes, and no one fully knows why, as first reported by The New York Times.

Just last month, Cursor, a coding tool, had to calm down angry customers after its A.I. support bot wrongly told them they could no longer use the product on multiple devices.

“We have no such policy. You’re of course free to use Cursor on multiple machines,” CEO Michael Truell clarified on Reddit, blaming the bot for the false message, as reported by The Times.

Advanced A.I. systems from OpenAI and Google and DeepSeek in China experience increasing occurrences of “hallucinations” which are errors in their operations. The tools use their “reasoning” abilities to solve problems but they frequently produce incorrect guesses and fabricated information.

The Times reports that the results of new model testing revealed that these systems generated fabricated answers in 79% of their responses.

“Despite our best efforts, they will always hallucinate,” said Amr Awadallah, CEO of Vectara and former Google executive, as reported by The Times. The bots operate without set rules because they produce responses based on probability which leads to fabricated information.

That’s a big issue for users handling legal, medical, or business data. “Not dealing with these errors properly basically eliminates the value of A.I. systems,” said Pratik Verma, CEO of Okahu.

In one example, AI-generated errors are causing real-world problems, especially in sensitive areas like legal work. Lawyers have faced sanctions for using fabricated information from AI models in court documents.

A report revealed that two lawyers in Wyoming included fake cases generated by AI in a lawsuit against Walmart, resulting in a federal judge threatening sanctions. This has triggered warnings in the legal field about the risks of relying on AI for tasks that require verified information.

OpenAI’s o3 model produced hallucinations during testing at a 33% rate which was twice as high as the o1 model. The o4-mini model demonstrated the highest hallucination rate at 48%.“We are actively working to reduce the higher rates of hallucination,” said OpenAI spokesperson Gaby Raila, as reported by The Times.

These issues are compounded by concerns over AI’s impact on journalism. A study by the BBC found that popular AI chatbots struggle with news content accuracy, with 51% of responses containing significant errors, including fabricated quotes and factual inaccuracies.

Researchers say part of the issue is how these bots are trained. “We still don’t know how these models work exactly,” said Hannaneh Hajishirzi of the University of Washington, reported The Times.