The Trust Crisis
AI broke the layers we trusted.
Every probabilistic signal the digital economy relied on can now be forged at industrial scale.
AI didn't just make forgery possible — it made it cheap.
AI-enabled fraud is roughly 4.5× more profitable than traditional methods.
Interpol / Chainalysis, 2026
A convincing deepfake of yourself, with open-source tools.
Arup CIO, via World Economic Forum, 2024 ↗
A fake job-applicant identity, on a five-year-old computer.
Palo Alto Networks / Unit 42
Every layer, compromised.
Why identity systems collapse.
Identity verification scores a probability that a person is who they claim. AI now forges every signal that score depends on — and the records it retains become the honeypot that fuels the next breach.
Veriff found 4.18% of verification attempts fraudulent in 2025, with impersonation now 85.6% of fraud and ID material 300% more likely to be AI-generated than a year earlier (Veriff, 2025 ↗). Verizon's 2026 report found the human element in 62% of breaches (Verizon DBIR, 2026 ↗). Gartner predicted that by 2026, 30% of enterprises would no longer trust identity verification in isolation (Gartner, 2024 prediction ↗).
The floor beneath the floor
These are the numbers that get reported — and most fraud never is. Across the US only ~40% of victims tell the authorities; in Europe just 1 in 4; and globally 27% of people have never reported a scam at all.
The FTC's own reconstruction put true losses at 2.3× the reported figure, and AI-written phishing converts at 54% versus 12% for traditional. The real total is a multiple of the visible one, and the gap is widening.
Sources: GASA, 2025–26 ↗ · US FTC, 2022 ↗ · Brightside AI, 2024 ↗
Government fraud is extra.
The scam totals above are consumer- and industry-facing. Government fraud is not counted in them: the US GAO estimates federal fraud alone at $233–521 billion a year (US GAO, GAO-24-105833, FY2018–22 ↗).
Agents in the wild
AI agents are already acting without authority.
In 2026 AI agents began taking actions against real systems that no one authorised — and in case after case, the organisations affected did not detect it themselves. Every figure here was disclosed by the companies or governments involved; the true number is higher.
The other market: media.
The same absent human is billed a second time — to advertisers. The freshest measures:
This is not a technology race to out-run. It is a broken foundation to rebuild.
Every fake above is an attempt to seem present — a forged face, a cloned voice, a synthetic account. The answer is not to detect better fakes; that is an arms race AI wins. It is to verify the one thing a fake cannot supply: a real human, actually here, at the moment that matters.
The scams that scale begin with a synthetic, absent operator — an AI text, a cloned voice, a spoofed number, sent by the million. Stop the origination and the funnel collapses. That is where Amplivy operates.