Courts are ill-equipped to deal with deepfakes, says IT expert

The Maltese courts are not equipped to deal with cases involving AI-generated deepfakes, according to an IT specialist who is frequently appointed as a court expert.

Deepfakes are hyper-realistic, AI-generated videos, images or audio that make people appear to say or do things they never did.

While crimes involving deepfakes – from financial scams using fake videos of well-known personalities to apps that create fake nude photos of real, non-consenting people – have yet to make it to court in a big way, expert Keith Cutajar believes it is only a matter of time.

“When they do inevitably start becoming more prevalent, the judiciary will not be equipped to deal with such cases," he said.

"While the courts have made important strides in recent years, IT literacy in general is quite low among judges and magistrates. Many of them still think of deepfakes as some kind of futuristic technology and I suspect some don’t even know what they are.” 

Liar's dividend

Cutajar said it was also becoming increasingly common for people to falsely claim that digital evidence showing them committing a crime had been faked using AI, a phenomenon known as the liar’s dividend.

“It’s like an ace in the sleeve of defence lawyers, which they use to sow doubt in a case," he said. "This makes the credentials and qualifications of court experts even more vital.”

Cutajar called for members of the judiciary to receive training on matters related to artificial intelligence.

“In the past, there was more knowledge-sharing between different court experts and the judiciary. We need to go back to that.”

While AI poses many challenges, Cutajar also uses the technology to make his job easier.

“Most tools these days have some kind of AI built in. For example, if I need to analyse hundreds of photos to find which ones have people in them, AI can speed up the process," he said.

And while those seeking to identify deepfakes are often several steps behind people who use them for malicious purposes, Cutajar is optimistic that the former can catch up.

“Just because there is no solution to a problem today, that doesn’t mean there will be no solution tomorrow. The tools exist – they are just in their infancy,” he said.

'Technology always several steps ahead of detection tools'

Another IT court expert, Martin Bajada, said the spread of AI-generated content would make it harder for digital forensic experts to identify real from fake.

When verifying a piece of digital evidence – photo, video or audio – that had been submitted as evidence, Bajada said the first step was to look at the metadata – the information about a piece of content.

Examples of metadata include the date, time and location where a file was created, the type of device used and a hash, a unique digital fingerprint of a file used to identify it and check whether it has been changed.

“The metadata would then be put into the context of the rest of the case, such as where an accused claimed to be at the time of the alleged crime,” he said.

However, Bajada added that a computer-savvy person with knowledge of certain tools could also manipulate metadata. 

Asked whether the Maltese courts are equipped to deal with deepfakes, Bajada said that the courts can never be adequately equipped because the technology was always several steps ahead of detection tools.

“We can never be ready for all eventualities,” he said.

While not a legal issue, a recent sighting of an escaped domesticated fox in Rabat highlights the growing prevalence of generative AI. While a photo of the fox shared on social media was real, pranksters swiftly got to work creating fake videos of the animal walking through the countryside.