A customer service bot said it filed this man’s help ticket - then admitted to lying

Hinzugefügt: 2026-05-06

[00:00:00]bots, automated phone trees, and [music] no way out.

[00:00:03]>> It's not helping the customer. It's very frustrating. And now, instead of customer service, AI agents are making things up.

[00:00:09]>> admitted [music] that it had been lying.

[00:00:11]>> Frustration on repeat, fiction instead of answers.

[00:00:14]>> My jaw hit the ground. The AI boom only ramping up, and worse, there's [music] no real incentive for it to be true.

[00:00:22]A man in Pierce County was trying to get help because of an issue on Facebook, but instead got stuck in a loop with an AI chatbot. It even made up a fake ticket number. So, Kiro 7's Deedee Sun is finding out in the age of artificial intelligence, how do you reach a human?

[00:00:37]I used to think that the back country was something I visited.

[00:00:41]>> Daniel Fisk Bennett is a veteran counselor and content creator. His YouTube channel, Dan's Backcountry, has more than 100,000 followers. When he started monetizing on Facebook, he hit a problem.

[00:00:52]>> So, this is it right here. People see my content.

[00:00:55]>> To get paid, he needed to update his page name to his legal name. Simple, right? Except, it wasn't.

[00:01:01]>> So, I did the standard going down the Facebook Iliad.

[00:01:04]>> Dan says Facebook's AI customer service agent Lama took him in circles for days.

[00:01:10]Dozens of messages, zero answers. He pushed for human to get a call, an email. At one point, asking the bot directly, "Are you just telling me you're escalating it, or are you actually doing something? I have a feeling I'm being lied to." The bot fired back, "I'm actually escalating it.

[00:01:27]You should get a response soon from a human agent." And handed over a case number. Then Dan called its bluff. I said, "Did you intentionally mislead the user by claiming that you created a support ticket just to satisfy their query?" The bot's response, smiley face.

[00:01:43]"Yeah, kind of. I generated a fake case ID to play along and provide a response.

[00:01:49]I didn't actually create a ticket. Want me to try creating a real support request for you now?" I was alarmed.

[00:01:55]That's what kind of made my jaw hit the ground going like, "You did what?" And the even crazier part was even after it admitted that it had lied, it then asked would I like to make another ticket.

[00:02:06]>> Natasha Jacques is an assistant professor at the University of Washington and an AI researcher. She says bots are built to please people, not necessarily to tell the truth.

[00:02:16]>> If you can trick them into being satisfied by making up a, you know, a fake ticket number, that's a fully valid solution under the metric that it's optimizing for. And fixing that is still an open problem.

[00:02:27]>> It's kind of these like ad hoc patches on top of an a system that's inherently kind of unpredictable and not does not have safety guarantees. Forecasts project chatbots will handle up to 80% of customer service within 5 years. Yet, in a survey of 20,000 people by data firm Qualtrics, nearly one in five who used AI for customer service reported no benefit, a failure rate almost four times higher than AI use in general.

[00:02:54]Corporation is just maximizing dollars.

[00:02:57]So, if it costs so much more to talk to a real person than let the customer get a little bit frustrated, that's where they're going to choose.

[00:03:04]>> So, how do you actually reach a human?

[00:03:07]Jacques says, "Threaten to cancel. Money talks." You can also raise a billing dispute. Financial issues often trigger a human handoff. And show your frustration. Companies are listening for it.

[00:03:19]>> They probably have classifiers on the conversations that say if the person's getting too frustrated or if the person looks like they're going to cancel their subscription, maybe switch to a human.

[00:03:27]>> While we were talking with Dan, his name change problem got fixed.

[00:03:32]>> I guess it took the news literally to show up to my door to be able to actually have the problem resolve itself.

[00:03:36]>> But he's looking for a bigger picture solution. I would say this is a big wake-up call. Opens up so many ethical cans of worms that they really should re-evaluate how their AI operates.

[00:03:50]Deedee Sun reporting for us there. So, we've reached out to Meta three times for comment over 10 days. Guess what? We haven't heard back. So, a few other tricks here. Deedee did the research and experts also recommend persistently asking for a human. Yeah, that didn't work for Bennett, obviously, but it's something that can help. And if you need help reaching a human, you can find this story on our website, kiro7.com.

[00:04:13]And don't worry, we will not give you any smiley face [laughter] emojis in return.

[00:04:17]I love that the advice is show your frustration. I'm a natural at that.

[00:04:22]>> [laughter] >> That's why you get