- 23 hours ago
Category
📺
TVTranscript
00:05Dancing dogs and horny fruits.
00:08You're actually kind of fun to be around, banana boy.
00:11Just some of the videos made using artificial intelligence
00:14flooding social media.
00:18While some make us laugh,
00:21others have a darker motive.
00:24They want prosecuted.
00:25They're taking money under false pretenses.
00:27AI is making it easier to scam us.
00:30It was my voice. It was my phone number.
00:33On that day, I hadn't spoken to it.
00:36And trick us.
00:38I think that's human.
00:40That one is AI.
00:41Oh, wow, it's really hard.
00:44Advances in AI are making it easier to sow division
00:47and harder for us to tell fact from fiction.
00:51Do you remember the label of me?
00:53The video was doctored to say that I said swear words
00:57and then a racial slush right after it.
00:59In the age of artificial intelligence,
01:02Panorama asks,
01:03is seeing still believing?
01:22This is a story about trickery, fraud and deception.
01:41Well, that was fun.
01:42But of course, it isn't real.
01:45These shots of me arriving at work were made using the latest form
01:49of artificial intelligence.
01:51It's called generative AI.
01:53And it can create realistic videos and images
01:56from simple text descriptions.
02:00It's cheap, easy to make and looks amazing.
02:03But with little regard for the truth,
02:05I'm not that good at flying.
02:11Here's Johnny.
02:15Videos like this are all over social media.
02:19Easily produced, high volume and often very weird.
02:23They've even got a name, AI slop.
02:26Word of the year in 2025.
02:29You may have seen some.
02:31Like the adventures of Ronaldo.
02:34Give me Messi back now.
02:36And Lionel Messi.
02:37I had a dream.
02:38He's coming back.
02:39Mr. Noodles, enjoying his best life.
02:43Welcome back to Fruit Love Island.
02:46And Fruit Love Island.
02:48You're actually kind of fun to be around, banana boy.
02:51And you've no idea how popular videos of fruits cheating
02:54on each other have become.
02:57These videos are competing for our eyeballs.
03:00Don't worry, my dear.
03:02As we watch, the social media companies run adverts,
03:05making money for them and those posting.
03:11Analysis of 15,000 YouTube channels found AI slop
03:15had 63 billion views last year,
03:19earning the creators more than £90 million
03:23and helping the social media companies reach record profits
03:27of more than £240 billion.
03:31It's big business.
03:35So what's the problem if it makes us laugh?
03:39Isn't it just cute animals doing harmless things?
03:56Hello everyone.
03:57Today I want to briefly talk about a product
04:00that has been showing up a lot on the internet called Wuffy,
04:04the interactive robotic puppy.
04:06It walks, reacts to touch, wags its tail, and makes sounds.
04:13I know, Wuffy, right? Sounds a bit ridiculous.
04:16What harm is Wuffy going to do?
04:19But actually this dog was after more than our attention.
04:22Just before Christmas,
04:24Wuffy started showing up on social media feeds.
04:28Advertised as a toy of the year,
04:30it claimed to be powered by advanced AI.
04:37Marie Hunter saw the ads.
04:39She was smitten.
04:46Hello.
04:47Lovely to meet you.
04:48I'm Marianna.
04:48Nice to meet you, Marie.
04:50I'm Marie.
04:50Thank you so much for having me.
04:52Talk to me about when you first spotted Wuffy.
04:55It was on Facebook.
04:57An advertisement came up for the cutest little dog you've ever seen.
05:00I'd been after a dog, but because I've got mobility issues,
05:04I thought I can't walk it.
05:06So that would be perfect because it said it reacted to your voice
05:12and commands that you gave it.
05:14And I thought that's the perfect little animal for me.
05:17It was the answer to my prayers.
05:19It was the AI that I wanted.
05:23Woofie is the first robot puppy with truly adaptive behaviour,
05:27thanks to its advanced built-in artificial intelligence.
05:30Do you know how much it cost?
05:32It was, I think, $49.99,
05:34but with it being a Christmas present,
05:36you don't really ask how much things cost.
05:39Shall I go and get him?
05:47This is Woofie.
05:54I remember buying one of those for my children 40 years ago.
06:03There's no AI involved in that dog.
06:07Before buying Woofie,
06:08Marie checked out the customer reviews online.
06:12And today I'm going to tell you everything you need to know about Woofie Robots.
06:15And was this the kind of thing that you found quite persuasive?
06:18Yes.
06:18Whether it's the pros, the cons, and whether it's really worth your money.
06:22She's so believable.
06:23Yeah.
06:23And did you seek these reviews out?
06:25Yes.
06:26To check?
06:26Yes, I wanted to have a look what other people thought of them,
06:30and they were all positive.
06:31But those reviews weren't what they seemed.
06:37She's AI-generated.
06:38I'm shocked.
06:42I wouldn't have thought that was possible.
06:45I've been fooled again.
06:47That's how I feel with something like that.
06:50Gullible.
06:51Part of my job is spotting AI fakery.
06:54So I know the signs to look for, like a monotone voice or strange eye movements.
06:59It can be subtle and difficult to detect.
07:03This isn't some cheap plastic toy.
07:05This was an elaborate scam.
07:07A fake AI dog with fake AI advertising and fake AI reviews.
07:13There have been thousands of unhappy customers commenting online.
07:17This is supposed to be an AI dog.
07:20Look at Woofie.
07:22This dog don't do nothing but bark, man.
07:28This is all he does.
07:32Woofie is labelled Made in China and sold by a company in Lithuania.
07:37The Advertising Standards Authority told them to stop.
07:41The company apologised.
07:43But the ad is still online.
07:49Marie felt embarrassed that she'd been caught out.
07:51But actually, I don't think she should be.
07:53It could happen to any of us.
07:55Increasingly, we're recommended AI content on our social media feeds.
07:58We look at it on small screens.
08:00We see it again and again.
08:01It's kind of easy to be fooled.
08:10Because AI-generated video is making fake content harder and harder to spot,
08:15people are becoming less confident in their ability to question it.
08:19If I was to tell you that the benchmark for measuring how AI video has evolved
08:24involves a bowl of spaghetti and the actor Will Smith,
08:27I think you'd think I'd lost it.
08:32In 2023, this was the best AI could do to make Will Smith look like he was eating spaghetti.
08:39Cartoonish, not at all believable.
08:42Two years later, it looked like this.
08:48And now, in 2026, incredibly realistic.
08:53Man, we've come a long way.
09:04Professor Yvonne McDermott-Reese specialises in AI and has concerns about the difficulty spotting it.
09:13The more that people see AI content, the less able that they are to discern fact from fiction,
09:20then the more likely they're going to be to distrust real content.
09:24Are we good at spotting fakes?
09:26The research suggests that we're not much better than chance.
09:28It's around 55% accuracy level.
09:31But what's interesting is that people aren't very good at spotting deep fakes,
09:37but they think they're a lot better than they actually are.
09:40Yvonne uses a test to show how difficult it's become to tell real faces from artificial ones.
09:46I'm going to show you 20 faces and it's up to you to decide whether you think they're human or
09:51AI-generated.
09:52OK.
09:53So here's the first one.
09:55Smiley lady.
09:59I think that's AI.
10:00You're correct.
10:05I think that's human.
10:06You're correct.
10:11Human?
10:12That one is AI.
10:14Oh, wow, it's really hard.
10:16It's almost like the AI looks better than the human beings.
10:19Yes.
10:21Early AI faces often had flaws like distorted teeth or ears that didn't look right.
10:27Now they perfectly match the human face, and for me it makes telling them apart really tricky.
10:3614 out of 20.
10:38I'm happy with that.
10:39Yeah, well done.
10:40That makes you a super AI face detector because the average person scores about 11 out of 20.
10:47So the bar is pretty low.
10:50The broader point is this.
10:51It shouldn't fall on just the ordinary person to have to try and figure out what's real and what isn't.
10:57It should be on the tech companies that have unleashed the technology through which fake images and videos can be
11:04generated at speed and at scale.
11:11We've got to a point where AI can generate realistic faces.
11:17So what can the technology now do for voices?
11:22There's been an uptick in highly convincing scams that deep fake people's voices, so it sounds like you're being called
11:29by someone you know, or at least someone you think you know.
11:34Hello.
11:35Hello.
11:35Lovely to meet you.
11:36Mariana.
11:37Hi, Jade.
11:37Nice to meet you.
11:38Last year a member of Jade's family got a phone call.
11:42It sounded like Jade on the call, but it wasn't.
11:46What do you know about what was actually said on that call to your relative?
11:50It was my voice, and I was saying to her that I owe someone money, and I need to pay
11:57them now.
11:58I sounded distressed, sounding like I'm about to cry.
12:02Did they have any doubt at all that it was you?
12:05No.
12:06Because my name came up on their phone, and it sounded like me.
12:10There wasn't much questions asked.
12:12It was kind of like, okay, Jade's in a problem, a situation, she needs me, kind of thing.
12:18Jade thinks AI was used to clone her voice, fooling her relative into handing over a large amount of money.
12:26She was saying that they stayed on the phone the whole time, so it was like, yeah, it went through.
12:31Can you send some more?
12:32And then it was a bigger amount, and then it was like, can you send even more?
12:36That must be so sort of confusing to be told that that's happened.
12:40Absolutely.
12:40Especially when I'm being told it's my voice.
12:42It was my phone number.
12:43On that day, I hadn't spoken to her.
12:45I checked my call list.
12:46I hadn't spoken to her a week before or a week after.
12:50Jade didn't call the police, but persuaded her relative to call the bank.
12:55It said because the transfer was authorised, it wouldn't refund the money.
13:00It's like, where's your proof?
13:01However, it doesn't stay away from the fact that it was fraud,
13:05and, you know, banks and all these other institutions should be aware of it,
13:09you know, because it's easily done.
13:14AI voice deepfakes are one of the fastest growing scams.
13:19A survey suggests one in ten adults globally have been personally targeted by an AI voice deepfake.
13:27Olly Buckley is a professor in cybersecurity.
13:31He's an expert in how AI is used to clone voices.
13:36Anyone could do this.
13:37Now, you don't need a powerful laptop.
13:39Anyone could spend a few pounds a month and have access to things that would have blown Hollywood out of
13:45the water a few years ago.
13:45How have audio deepfakes changed in the past even 12, 18 months?
13:51They've changed significantly.
13:53They were robotic.
13:54The tone of voice didn't match the words that were being said,
13:56so someone might be saying something that's quite angry or aggressive,
13:58and it just sounded quite flat and monotone.
14:01Now it sounds like the person.
14:03It puts in the ums and the ahs and the pauses and the breaths.
14:06It's really impressive but also quite scary to separate what's real and what's not real.
14:12Olly is going to show me just how convincing it can be.
14:16I need you to talk for like 20 seconds and I'll record you talking.
14:19Let me hit record.
14:20Tell me when.
14:20OK.
14:22My name is Mariana.
14:23I investigate social media for my job at the BBC and for this panorama I'm looking into AI and deepfakes.
14:31It felt like the dog that hadn't barked and now it really feels like it has.
14:35OK.
14:36That might work.
14:36Let's try that.
14:38Hello.
14:39Hey, it's Mariana.
14:40Hope I'm not catching you at a bad time.
14:42Are you free?
14:42I am free.
14:43I wanted to check in with you about if we're still meeting for dinner later.
14:51So the lag is it waiting to see if it's quiet basically to see if you've finished talking.
14:56Oh, hey.
14:57Yeah, I was just thinking about that.
14:58I don't remember us making solid plans for dinner but I'm definitely open to it if you are.
15:03What were you thinking?
15:04I like how my AI voice is cancelling my plans.
15:08Yeah.
15:08It's so responsive.
15:10It really is.
15:10It knows how people talk and it knows how a conversation works and it's just reading it out in your
15:14voice.
15:20Fake AI content isn't just about fraudsters and scam artists.
15:26There's a new type that's trying to provoke and even influence how we think.
15:34Croydon in South London is known for many things.
15:37Surrey Street Market, Cape Moss and most recently an escalator.
15:42Croydon can lay claim to having the most famous escalator in the world after Taylor Swift filmed a recent video
15:49here.
15:56What you won't find here though is Croydon Water Park because it doesn't exist.
16:06Yet AI generated videos of Croydon Water Park have been taking over social media.
16:12They feature crowds of men, mostly black, wearing balaclavas and padded jackets.
16:18Slipping down a water slide into a dirty swimming pool.
16:23It's absurd.
16:24But it's part of a worrying and wider trend where online influencers are creating AI videos that portray London and
16:31other UK cities as being in a state of urban decline.
16:38These fake AI videos feature road men, a derogatory term for some young men.
16:44Attention mandem, your benefits have been paid.
16:47They are often shown collecting benefits.
16:52And behaving anti-socially.
16:55It's called decline porn.
16:59These Croydon videos have racked up more than 50 million views and are reaching people way beyond South London.
17:07Local people say they're having a negative impact.
17:12These Croydon AI videos need to stop.
17:14They are not funny.
17:15Croydon is not that bad of an area.
17:17He is not as ghetto as people think.
17:19But these videos are making people think this is real life.
17:22Leading the fight back is Corey Hilton.
17:25Born and bred here, he's a proud Croydonian.
17:28And he's fed up with these decline porn videos.
17:33At first I thought it was funny.
17:36But then I'm like, come on.
17:38This isn't what Croydon is.
17:40Croydon doesn't actually look like that.
17:41What do you think about this idea that Croydon is full of road men?
17:46It's a dumb stereotype to be fair.
17:48As you know, Croydon is very diverse.
17:52And what do you think about the impact of those kinds of AI fakes?
17:55For the younger generation, it's a bit of laughs, like, not taking it seriously.
18:01But for the older generation that are not really tech savvy, they're really thinking it's real.
18:07They're getting fooled by it.
18:09It kind of sets us back.
18:11It kind of makes people think this is what Croydon is.
18:13It's just black people in puffer jackets, walking around, talking slang, not articulating their words or stuff like that, which
18:21isn't the case.
18:25So, who's making these videos and why?
18:29I tracked down the creator to ask him.
18:33Not only does the person behind these videos not live in Croydon, they've never even been to the place.
18:40He calls himself Radial B and won't tell me his real name.
18:45He's agreed to speak to me, but only over the phone.
18:49Why do you do it?
19:04What would you say to someone if they were like, oh, actually, I don't like the videos very much?
19:12What would you say to someone if they were like, oh, actually, I don't like the videos very much?
19:19What would you say to someone who argues that that kind of content is racist, it's reinforcing racist stereotypes?
19:33Radial B is a new brand of online faker, one who seems to thrive off engagement, but shrugs off any
19:39responsibility for how that content can be weaponised.
19:48The Croydon videos are just the tip of the iceberg.
19:52There are hundreds of other decline porn videos that go further.
19:56And like Radial B, many people making this content have little or no connection to the place they are portraying.
20:05For the places themselves, this can have real-world consequences.
20:09Like in London, it seems like this kind of content is distorting or changing the perception of the city.
20:17Good evening, we're live over central London where a devastating explosion has ripped through the London eye.
20:23From AI videos showing London under attack to Parliament enacting Sharia law.
20:28From this day forth, the consumption and sale of pork and alcohol shall be haram and thus strictly forbidden within
20:33the United Kingdom of Brita Arabia.
20:36They focus on politically divisive topics like immigration or crime.
20:41Others imagine a future version of the capital.
20:49The mayor of London, Sadiq Khan, thinks his city is on the front line of the fight against this new
20:55kind of disinformation.
20:58There are consequences for these sorts of AI-generated lies.
21:02People overseas being scared to come to London.
21:05Students being scared to study in London.
21:07People not wanting to invest in London.
21:10We've done some research.
21:11What we've seen is over the last two years, there's been a 150% to 200% increase in narratives
21:21of London being in decline.
21:22What do you think the purpose of that kind of content is?
21:24One is you've got state actors. We've seen evidence of Russian activity, China's activity, MAGA activity from the USA, the
21:34extreme right wing.
21:35Secondly, we've seen individuals and companies trying to monetise and make profit from division.
21:44What would you say to someone who points out there are issues with crime in London?
21:49We're not perfect. There are all challenges in this great city of ours. We're dealing with them.
21:53And me taking on disinformation doesn't take away from me trying to take on the challenges that our city faces.
21:59How do you think London can counter this kind of disinformation?
22:03We in City Hall are working with the police and public health officials to fight back to debunk some of
22:08these myths and take on some of these lies.
22:10To give people the resilience and confidence to know that something is clearly a lie.
22:20The way AI is being used to spread disinformation is fast becoming one of the defining challenges of our time.
22:27Wherever it occurs, the consequences can be life changing.
22:34I was so paranoid of going out, especially after receiving the death threats.
22:38I didn't want to put myself in any danger.
22:43Cheryl Bennett was head of year at a secondary school near Walsall.
22:48In 2024, she agreed to help out a colleague who was standing for the Labour Party in the local elections.
22:56I said to him that I would help him go out and deliver some leaflets.
22:59But I said, don't ask me to do any speaking because I don't know anything in terms of political parties
23:05whatsoever.
23:06So, he was happy with that.
23:09I went to post the leaflet through the door and the gentleman then answered.
23:17I just asked, have you voted? And he said yes. So, I asked him then, who have you voted for?
23:23Who did you vote for? No problem, thank you.
23:26OK, thank you.
23:27And he said a name, but I couldn't understand what he had said.
23:30So, I was like, OK, no, it was lovely. Thank you. And, like, walked away.
23:38Cheryl thought no more about it, but a few days later, her nightmare began.
23:45My phone just started going crazy.
23:48I looked at my emails and it was off of this one parent saying, is this you?
23:53With loads of question marks.
23:55Cheryl discovered there was footage of her delivering the leaflet.
23:59It had been manipulated and was circulating on social media.
24:03That's when the abuse started.
24:05You're racist, you're a dirty little bitch, you're a slag, you're a rat.
24:09I didn't expect a teacher of your standards to be so racist.
24:12You know, it was all these comments and I was like, what on earth is going on?
24:16What did the doctored videos show and say?
24:19As I was walking away, it was edited to say that I was having a said swear word
24:24and then a racial slur straight after it.
24:26OK, thank you.
24:29That was shared on TikTok.
24:37But then before I knew it, it was all over Facebook.
24:39It was all over X.
24:41It was being shared everywhere.
24:43Do you know how many views overall that clip had in the end?
24:48I think it was over 2 million views across all different platforms.
24:52That is a big number.
24:53Yeah, huge.
24:55I have no words for this.
24:58The video was shared by Ahmed Yacoub, an independent candidate standing for West Midlands mayor.
25:05Those who are still with the Labour Party, now is the time to leave.
25:11He then shared my name, my place of work, people started saying what type of car I was driving.
25:17Was that scary?
25:18It was.
25:19And especially when the death threats started coming through.
25:22West Midlands police contacted Cheryl, but soon realised the footage had been doctored
25:28and that she was completely innocent.
25:30She sued Ahmed Yacoub.
25:32He settled the case, paid her an undisclosed sum and has said he regretted sharing the video.
25:38No one has been arrested or charged and Cheryl still doesn't know who created it.
25:44So in some ways you were like collateral damage in this political debate.
25:47And I think that's the part which also hurts the most because as a human being there was no concern
25:54for the long lasting effects that it would have on me or anything like that.
26:07So what can we do to navigate this new era of AI misinformation?
26:12Here in Cambridge, a team think they might have an answer.
26:17They've produced what they say is a vaccine.
26:21They call it psychological inoculation and today the team are taking it into Hills Road Sickform College
26:27and I've come to see the students get their jabs.
26:31Just as vaccines introduce a weakened version of a pathogen into the body, right,
26:34to try to trigger the production of antibodies to help confer resistance against future infection,
26:39turns out you can do the same with information.
26:42Professor of psychology Sander van der Linden has developed an award-winning game called Bad News.
26:49It reveals the techniques used to spread misinformation.
26:53He's been testing it nationwide.
26:57The game is called the Bad News game which is a pun and the goal of the game really is
27:01to pre-bunk manipulation techniques
27:03by letting students step into the shoes of an online manipulator.
27:09By deconstructing the manipulation tricks that misinformation producers use in advance,
27:13people can become more resistant to misinformation in the future.
27:16What do we know about the effectiveness of pre-bunking?
27:19Pre-bunking generally works better than debunking because you come before
27:23and so there's more of an opportunity to actually help people identify and neutralize it, if you will.
27:30How did you guys find the game? Enjoyed it? Didn't enjoy it?
27:35It kind of like offers you to like a bit more of an insight into like what exactly these people
27:41are doing to manipulate the population.
27:43It has like affected how much I trust things now because it's like okay well they're probably lying to me
27:50but like how am I ever going to know they're actually telling me the complete truth.
27:54The main social media companies have guidelines for how AI is used and labelled but that doesn't seem to be
28:01solving the problem.
28:01It can be beneficial but then it always depends on the intent of the person that's behind it.
28:08When stuff pops up on your feed are you more skeptical?
28:11More than ever. More than ever, yeah. And I tend to just scroll past them now.
28:16Once we can no longer believe our eyes anymore then that becomes really quite problematic.
28:22In a world where it's increasingly difficult to tell fact from fiction, the fear is we'll stop trusting anything we
28:30see online.
28:31What's wrong with you?
28:34What?
28:59When stuff picks up often?