Real enough

When Taylor Swift married Travis Kelce at Madison Square Garden in July 2026, guests reportedly signed non-disclosure agreements and surrendered their phones. More than two months later, fans are still waiting for a verified photo of the newlyweds.

It hasn’t stopped the internet trying to fill the gap. Images of the celebrity couple quickly appeared online - some obviously fake, others convincing enough to make fans look twice. The “leaked” photos appeared to have been taken secretly and were blurry or poorly framed. Swifties quickly got to work and declared what many suspected – the photos were fake and created using generative artificial intelligence (GenAI).

In this case, the stakes are low. The consequences of believing a fake celebrity wedding photo pale in comparison to believing a fabricated satellite image from a war zone.

Earlier this year, BBC Verify reported an unprecedented wave of AI-generated misinformation, including fabricated videos and satellite imagery, with fake content viewed hundreds of millions of times online.

Closer to home, AI is being used to make scams more convincing. In the 2026 financial year, the Australian Securities and Investments Commission removed more than 19,400 scam websites, with GenAI behind many of them. Scammers have used deepfake footage of well-known Australians including Anthony Albanese and Dick Smith to convince people to hand over their money.

It's not picture perfect

Associate Professor Michael Mehmet built the University of Wollongong’s (UOW) first generative AI subject in 2024. The marketing expert, from UOW’s School of Business, was convinced his students would otherwise graduate into a world their degrees had not prepared them for.

He’s clear about where image generation has landed.

“Honestly, with images, it’s really hard to tell these days,” Associate Professor Mehmet said.

“Before it was easy. If the person was too perfect or super good looking or too symmetrical, you knew it was fake. But you can prompt them to not look flawless now. I don’t judge anyone who gets tripped up by images.”

He’s seen the shift in his own classroom. In 2024, one student out of 127 produced what Associate Professor Mehmet described as a “half decent AI video”. Last year, about half could. And this year? Most of his students could, using tools available for free.

“A couple developed stuff and I couldn’t tell,” he revealed.

What still gives it away?

While images are increasingly hard to spot, AI-generated video is easier to identify because there’s more to get wrong.

“Video has transitions, people move, faces move, backgrounds change,” he said.

“The classic things that you can usually spot are the shading and the background, if there are logos or text in the video. It can still look a bit glitchy.”

The giveaways

Text and logos: Letters may dissolve or look warped, while logos are often wrong.

Physical distortions: Look for extra or fused fingers, jewellery that blends into skin, distorted teeth when someone is speaking, or objects that bleed into each other.

Light discrepancies: Shadows falling in the wrong direction, reflections that don’t match, or people in the same image with inconsistent lighting.

Audio drift: Lips that don’t quite match the words or the speed of the audio.

The technical checks

If you’re unsure, there are a few ways to investigate further.

Reverse image search: Use Google Images or TinEye to see whether an image has appeared elsewhere, or where it first appeared online.

Metadata and watermarks: Look for creation details or digital signatures. Tools such as SynthID can help identify AI-generated content.

And it never hurts to get a second opinion. Ask a friend, family member or professional if something doesn’t look right.

Better detection methods are also coming, including Content Credentials, which records how a file was made and edited.

But there’s a catch. Most online images and videos carry no obvious watermark or tell-tale sign. It’s why Associate Professor Mehmet argues we might be looking in the wrong place.

“People have to think about the context of an image or video. We must start thinking about all the things that surround it,” he said.

“We need to think about who is sharing it, where it’s been published, what text accompanies it, what reaction the content is trying to elicit from the viewer.

“If you put the tech lens and critical thinking lens over each other, you should be able to discern whether it’s AI, or at least pause and question its validity.”

The picture is different when you zoom in

While checklists may help, for Dr Karley Beckman, Senior Lecturer in Digital Technologies at UOW and Chief Investigator with the ARC Centre of Excellence for the Digital Child, the wider issue is digital literacy – and what sits beyond it.

“Digital literacy is the ability and knowledge required to use digital technologies,” Dr Beckman said.

“It includes not only knowing how to use digital tools, but also understanding when, why, and whether they should be used.”

But knowing your way around technology or an online environment isn’t the same as being able to judge what’s on it. This is a skill Dr Beckman refers to as media and AI literacy – it’s about understanding, evaluating and responding to information and content within those environments.

It’s a gap she’s working hard to close. Dr Beckman has led the development of two AI-focused subjects in UOW’s School of Education, designed to build teachers’ own AI literacy before they teach students how to navigate AI-saturated environments.

For young people her starting point is even simpler.

“Children need to assume that any content and media they encounter online could be AI generated,” Dr Beckman said. “Supporting the development of critical thinking is a skill that is increasingly important.”

“This includes skills to question, identify, and check the trustworthiness of not only the media, but the location or source of information,” Dr Beckman said.

“Media found on social media would need a much higher level of scrutiny than media found on a news website. So we need to be more judicious about where we consume media.”

But Dr Beckman is clear that the responsibility can’t rest with individuals alone.

“Expecting individuals to navigate AI-generated content without guardrails is an impossible task that generates many harms and risks to individuals,” Dr Beckman explained.

“What is needed is greater protections, through regulation, and accountability for technology companies that create products that facilitate the creation of harmful content and misinformation.”

Dr Beckman said the draft  Digital Duty of Care Bill, is a step in the right direction. The Draft Bill, released in September 2026, targets some of the most harmful instances of GenAI images. The onus will be on providers to keep users safe from harm when using their platforms. Dr Beckman is among researchers from the School of Education who are preparing a response to the Draft Bill.

Back to the future

GenAI is evolving at a rapid pace and Associate Professor Mehmet has no doubt about the next five years.

“You won’t be able to tell the difference between an image and reality.”

Which means that for those of us trying to work out what’s real and what’s not, we may need to look beyond the picture itself.

“It’s not what you see that you trust, it’s about who says it,” Associate Professor Mehmet explains. “It goes back to source believability.”

Like Dr Beckman, Associate Professor Mehmet suggests thinking critically about where you saw the content, and who shared it with you. Check whether a trusted news source has independently reported the story. And perhaps, most importantly, slow down before sharing.

Artificial intelligence is here to stay. We must move from simply learning to use the technology to navigating a digital world where seeing is no longer believing. It’s clear that digital literacy is more important than ever.

So is building the people capable of helping the rest of us navigate what comes next. And UOW is preparing students for that on two fronts.

UOW's education courses are preparing the next generation of educators to teach with, and about, AI technology. As Dr Beckman said, they're the ones who will pass these vital skills on. The Bachelor of Computer Science spans artificial intelligence and big data, cyber security and digital systems security. It's about the skills behind the detection and verification tools we will all come to rely on. And

But perhaps the most important won’t be knowing exactly what AI can do. It will be knowing when to stop, look again, and ask: is this really what I think it is?