Every Image Is Guilty Until Proven Authentic

A retiree in Saskatchewan handed over three thousand dollars to someone she believed was Prime Minister Mark Carney. She watched a video of him endorsing a cryptocurrency investment. His face, his voice, the C.B.C. logo in the corner — all of it was fake.

That woman isn't careless

That woman isn't careless. She saw what looked like a credible news broadcast featuring her own prime minister. And she's far from alone. According to Canadian authorities, victims lost more than three hundred and eighty-eight million dollars to crypto scams between January 2024 and September 2025. Officials estimate only about one in ten to one in twenty victims ever report it. If you've ever watched a video online and assumed the person in it was real — this story is about you. Deepfake technology — A.I. that generates synthetic faces, voices, and video so realistic they fool trained professionals — isn't a future threat. It's already embedded in fraud at scale. Investment scams, identity theft, political manipulation, even fabricated evidence in criminal cases. So the question running through all of this: if any image or video can be manufactured, what does it actually take to prove something is real?

Start with the numbers, because they tell you how fast this moved. According to industry data compiled by Fourthline, deepfake-driven fraud losses topped four hundred and ten million dollars in just the first half of 2025. Projections put A.I.-enabled financial fraud at roughly forty billion dollars a year by 2027. That's not a slow creep. That's an explosion. And it's not just about money disappearing from bank accounts. It means the systems banks and fintechs use to verify your identity — the ones that ask you to hold up your driver's license and take a selfie — those systems are now under direct attack.

According to research from HyperVerge, deepfake attacks now cause one out of every twenty identity verification failures during account signups and transactions. One in twenty. That means someone submitting a synthetic face — a face that doesn't belong to a real person — gets past the front door five percent of the time. For investigators and compliance teams, that rewrites the playbook on onboarding risk. For the rest of us, it means someone could open a bank account using a face that looks like yours but isn't.

The Saskatchewan scam used a specific playbook

The Saskatchewan scam used a specific playbook. Fraudsters generated a deepfake video of Prime Minister Carney, layered in C.B.C. branding to make it look like a real news segment, and pushed it through social media ads. The target — a retiree — had no reason to doubt what she was seeing. That same pattern is showing up everywhere. Attorneys general in both Illinois and New York have issued warnings about deepfake investment scams running on Meta platforms. The United Nations published a report calling weaponized A.I. fraud a global wake-up call, pointing to billions in illicit financial flows tied to synthetic media.

Now, the investigative side of this gets technical, but it matters. Traditional facial recognition asks one question — who is this person? That's no longer enough. Investigators now need a second question answered first — was this face ever synthetically altered? That means examining compression artifacts in the image file, checking whether light reflections in both eyes match, and analyzing whether facial movements stay consistent frame by frame. If you've ever noticed something slightly off about a person's face in a video — a weird shimmer around the jawline, eyes that don't quite track together — that's what forensic analysts are looking for, just at a much deeper level. The teams getting ahead of this are building that forensic check into their first step, not their last.

And the voice side is just as urgent. According to data compiled by Axis Intelligence, the number of deepfake files online jumped from roughly half a million in 2023 to eight million by 2025. A.I.-generated voices have crossed what researchers call an indistinguishable threshold — meaning human ears can no longer reliably tell the difference. Major retailers report dealing with more than a thousand A.I.-generated scam calls every single day. A thousand a day. That means the voice on the other end of a phone call carries roughly the same evidentiary weight as a stock photo — almost none, without verification.