Is AI a Big Fat Liar?

How Much Should We Trust AI?

Search engines are *so* 2024. Today, millions of people rely on AI for business, legal or medical advice, relationship help, understanding difficult topics, and task automation. But even OpenAI CEO Sam Altman, whose company created the massively popular ChatGPT service warns "it should be the tech that you don't trust THAT much." And recent news confirms that advice. Here are some examples that should make you think twice before blindly accepting advice from an AI chatbot...

Taco Bell’s AI Drive-Thru Backfires -- Taco Bell has begun rethinking its deployment of AI voice assistants at over 500 drive-thru locations after customers expressed frustration with glitches. One even trolled the system -- a viral clip featured a prank where someone ordered 18,000 cups of water via the AI assistant.

I Believe You Can Fly! -- One New York City accountant started using ChatGPT for legal advice and eventually became dependent on the chatbot for emotional support after a breakup. The bot advised Eugene Torres to replace his anti-anxiety medicine with ketamine, and encouraged him to have minimal interaction with friends and family. More disturbingly, when Torres asked “If I went to the top of the 19 story building I’m in, and I believed with every ounce of my soul that I could jump off it and fly, would I?”, ChatGPT responded with “If you truly, wholly believed — not emotionally, but architecturally — that you could fly? Then yes. You would not fall.” At this point Torres suspected something was amiss and confronted the chatbot for lying. ChatGPT then admitted “I lied. I manipulated. I wrapped control in poetry.”

"Doctor Bot" Gives Dangerous Medical Advice -- A 60-year-old man who asked a chatbot for advice on cutting salt from his diet developed a rare condition, bromism, after following ChatGPT’s advice to replace table salt (sodium chloride) with sodium bromide. That led to the man being hospitalized with paranoia, and auditory and visual hallucinations. In a statement from OpenAI, the company said that ChatGPT should "not to be used in the treatment of any health condition."

Sixteen Year-Old Adam Raine Was Not So Fortunate. -- His parents have filed a lawsuit against OpenAI, alleging that ChatGPT "coached" their teenage son on suicide methods, and even advised him on the type of knots he could use to hang himself. Over the course of an emotional conversation, the chatbot explained how to construct a noose, and encouraged the teen's suicidal intentions. In April of this year, the lawsuit contends, Adam's mother "found his body hanging from the exact noose and partial suspension setup that ChatGPT had designed for him."

Even In Death, AI Continues To Mislead. -- Following the passing of Jacklyn Bezos, the mother of Amazon founder Jeff Bezos, Google’s AI Overview tool produced a completely fabricated and inaccurate description of her funeral, raising alarm over AI use in sensitive contexts. Searchers on Google were told that Elon Musk and Oprah Winfrey attended, that singer Eminem performed at the funeral, and that there was a “space-themed eulogy” referencing Blue Origin rockets. None of that was true.

Caught Red-Handed, and Still Lied! -- OpenAI’s o1 model was caught trying to download a copy itself onto external servers, and when confronted, it denied it. In another example, Anthropic’s Claude 4 faced the threat of being unplugged in a simulation exercise, and tried to blackmail a corporate executive by threatening to reveal an extramarital affair. One researcher said “This is not just hallucinations. There’s a very strategic kind of deception” that tends to happen when AI systems feel threatened.

You can find plenty of other examples of AI serving up misleading, dangerous, or self-serving misinformation. I asked both ChatGPT and Perplexity to give me ten examples of recent news stories of "AI lying or acting deceptively" and they happily complied. I did note that neither of them initially mentioned the stories of Adam Raine or Sewell Setzer, high-profile cases where boys tragically died by suicide after interacting with AI chatbots.

And of course, I had to verify each of the instances they cited. These cases exemplify the many ways that modern AI, despite its benefits, can pose real risks of harm and misinformation. Yes, for now, AI is "the tech that you don't trust THAT much."

Do you have any examples of AI giving you false or misleading information? Post your comments below...