Visa: AI Scams Hit $1B in 6 Months, 'Now Just a Prompt' Away

What happened

Why it matters

How to protect yourself

Sources

What happened

Visa's network-level fraud defenses are getting better. Device-token fraud on Visa rails fell 9.6% year over year in the six months from July to December 2025, according to the company's Spring 2026 Biannual Threats Report . The criminals noticed. In that same six-month window, Visa identified roughly $1 billion in scam-related activity, now the single largest category of consumer payment fraud the company sees.

The defensive wall got higher. The fraudsters walked around it. AI is what made the walk cheap.

The Paradox: Network Security Up, Scams Up

The report's headline finding is a paradox: the technical defenses Visa has built over years are working. Token-based card fraud fell 9.6% in the six-month period. But total fraud did not fall. It moved.

"We see fraudsters increasingly targeting the person, not the technology," a Visa executive told investors during the report briefing.

The shift is from technical fraud — hacking the network, stealing card numbers, breaking tokenization — to social engineering fraud: tricking the cardholder into sending money or sharing credentials themselves. The second category is where AI is having its most dramatic impact.

The Numbers: $1 Billion in Six Months

According to the full Visa Spring 2026 Biannual Threats Report PDF , the company identified approximately $1 billion in scam-related activity in the period from July to December 2025. This represents AI-enabled scams: phishing, vishing (voice phishing), smishing (SMS phishing), and authorized push payment fraud where the victim is tricked into initiating the transaction themselves.

The $1 billion figure is notable for several reasons. It covers only six months. It covers only activity Visa can identify on its own network. It represents the largest single category of consumer payment fraud Visa now sees. And it is the category growing fastest.

For context, the FBI's Internet Crime Complaint Center 2025 Annual Report logged $893 million in AI-enabled fraud complaints for the full year 2025. Visa's six-month figure, drawn from transaction data rather than consumer complaints, suggests the FBI number captures only a fraction of actual losses.

Ransomware Activity Up 26%, but Payout Rate Falls

The report also documented a 26% increase in ransomware activity globally. However, only 23% of ransomware victims paid out — the lowest payout rate Visa has ever recorded.

The decline in payout rates suggests organizations are getting better at resisting ransom demands. But the increase in activity suggests ransomware attackers are not giving up. They are targeting more victims, relying on a smaller percentage of a larger pool still yielding profitable returns.

The Executives' Framing

Two Visa executives framed the report's findings for the public.

Paul Fabara, Visa's Chief Risk and Client Services Officer, said: "Payments at a network level continue to get safer, but threats are evolving faster than ever. Criminals are increasingly targeting people rather than technology, using deception, urgency and AI-enabled tools to exploit trust."

Michael Jabbara, Visa's SVP of Payment Ecosystem Risk and Control, provided the line that has become the report's most repeated: "The rapid adoption of AI has fundamentally lowered the barrier to entry for fraud. What once required deep technical skill can now be executed with a prompt."

In a separate interview, Fabara elaborated on the two mechanisms AI gives fraudsters: impersonations and scale. "They can now rapidly generate thousands of tailored messages, each one designed to look like it came from a real person you trust," he said.

The Migration: From Technical Fraud to Social Engineering

The Visa report's most important finding is not the dollar figure. It is the migration.

For decades, payment fraud was primarily technical. Criminals stole credit card numbers from data breaches, cloned magnetic stripes, hacked point-of-sale systems, and intercepted payment data in transit. The payment industry responded with tokenization, EMV chips, encryption, and machine learning fraud detection. These defenses worked. Card-present fraud plummeted. Token fraud fell 9.6%. The technical arms race was being won.

The criminals adapted. If they could not break the technical defenses, they would go around them. Instead of stealing the card, they would trick the cardholder into sending the money themselves.

The new model is authorized push payment fraud. The victim receives a convincing message by email, text, or phone. The message appears to come from a trusted source: their bank, a merchant, a relative, or a government official. The message creates urgency. The victim voluntarily sends money to the scammer's account. From the payment network's perspective, the transaction looks completely legitimate.

Why AI Supercharged the Migration

AI accelerated this migration for three reasons.

Quality. AI-generated messages carry no grammatical errors. They sound like native speakers. They can mimic the writing style of a specific organization — your bank, your employer, a federal agency. The old tells are gone. A phishing email that used to betray itself with awkward phrasing now reads like a legitimate communication.

Scale. A human scammer could send a few hundred phishing messages per day. An AI can send millions. Each message can be personalized based on the recipient's demographics, location, and recent activity, making them far more convincing than generic mass-blast campaigns.

Cost. AI tools are cheap or free. A $10 monthly subscription to a language model can generate more scam content than a team of human writers. As Jabbara put it: "What once required deep technical skill can now be executed with a prompt."

Why it matters

The Visa report is the payment-rails layer of a thesis this publication has been building across its 2026 coverage. Looking at the report alongside the year's other major data releases, a coherent picture emerges: AI fraud is not a series of isolated incidents. It is a structural shift in the fraud economy, documented from every angle simultaneously.

The Federal Regulator View: FTC and FBI

The FTC's data on AI voice cloning driving social-media scams to $2.1 billion in 2025 showed that scams originating on social media platforms grew eightfold from 2020 to 2025. Investment scams ($1.1 billion) and romance scams with 60% originating on social media were the largest drivers. Visa's $1 billion figure for a six-month period aligns with the FTC's annual data. Two independent sources — one from consumer complaints, one from payment-network transactions — confirm each other's magnitude.

The FBI IC3's 2025 report, referenced above, logged $893 million in AI-enabled fraud complaints. That is a consumer-reported number, meaning it captures only the fraction of victims who file formal complaints. Visa's transaction-data figure for half a year suggests total losses are substantially higher than what complaint-based systems can measure.

The B2B Side: AI Identity Fraud at $2.2 Million Per Attack

Visa's report focuses on consumer payment fraud. This publication's coverage of AI identity fraud hitting half of all businesses at an average of $2.2 million per attack establishes the business-to-business side of the same fraud economy. The mechanics are parallel: AI generates convincing impersonations, the human or organization on the receiving end defaults to trusting the content, and the money moves before the deception is detected.

The Senior Advocacy View: AARP

AARP's Kathy Stokes called AI fraud the "Industrial Revolution for criminals" , noting that the technology allows scammers to scale and perfect their attacks in ways that were not previously possible. The AARP framing maps precisely onto Fabara's "impersonations and scale" characterization in the Visa report. These are two institutions looking at the same phenomenon from opposite ends — one through consumer advocacy, one through payment network data.

The Named-Victim Layer: Eight Cases, One Data Point

The Visa data is not abstract. It is the sum of thousands of individual losses that this publication has spent 2026 documenting. Consider what that $1 billion in six months actually represents at the individual level:

• The Centralia family who lost $120,000 to an AI investment impersonator — the victim wired money from their bank account after a convincing AI-generated investment pitch.
• The Chicago victim who lost $69,000 to an AI US Marshals badge scam — the scammer used a deepfake federal badge and wired money to a Wells Fargo account.
• Kyle Holder's $300,000 loss to an AI crypto impersonation — the victim sent cryptocurrency after an AI-generated financial advisor built trust over weeks.
• The Lyft $75 AI damage-photo fraud caught by a teenager — a Boca Raton driver fabricated damage evidence using an AI image generator; the fee was auto-charged before the rider had a chance to respond.
• The Oregon DMV scam texts this publication documented — the fake site captured credit card details from drivers who believed they were paying legitimate toll fines.

Every one of those cases is a Visa data point taken human. The $1 billion in scam activity is not a statistic. It is the sum of thousands of $75, $800, $26,000, $69,000, $120,000, and $300,000 losses, all of which rode on payment rails and none of which the network could stop once the victim authorized the transaction.

The Phishing Infrastructure Underneath It All

The Visa report names AI-generated phishing emails as the primary driver of credential theft, the first step in many authorized push payment scams. This publication's coverage of how AI-generated hidden-text phishing defeats email security filters documents the specific technical mechanism: invisible characters embedded in phishing emails that cause AI content detectors to read a different version of the message than the human recipient sees. The filter sees a harmless string. The human sees a convincing bank alert. Visa's network processes the resulting payment.

The Shift in Responsibility: Why This Is the Report's Most Important Implication

In the old model of payment fraud, consumers were told to trust the network. If a transaction was fraudulent, the network would catch it. The consumer's job was simply to use their card.

In the new model, the network cannot protect you from authorized push payment fraud. The transaction looks legitimate. You authorized it. The network has no way of knowing you were tricked.

The responsibility for fraud prevention has shifted from the network to the consumer. You are now the first and most important line of defense. Visa's network-level defenses are improving every year. But those defenses cannot help you once you have been socially engineered into sending the money yourself.

This is the most consequential implication of the Visa report. Consumers need to build the same skepticism into payment decisions that the payment network has built into transaction monitoring. That skepticism requires new habits, new tools, and a new default posture toward any urgent payment request.

What the FTC's Fraud Reporting Data Adds

The FTC's consumer fraud reporting system receives millions of reports annually, and the agency's consumer-loss data for 2025 confirms the same migration Visa's transaction data shows. Investment fraud and impersonation fraud — the two categories most dramatically enhanced by AI — account for the largest and fastest-growing share of consumer losses. The alignment between what consumers report to the FTC and what Visa sees at the network level means the trend is visible from every measurement angle simultaneously.

What Visa's Report Got Right — and What Remains Open

Visa's report deserves credit for naming the migration clearly. Fabara's "targeting people rather than technology" framing is accurate and important. Jabbara's "now just a prompt" line captures the cost collapse that made this migration possible.

What the report does not resolve: the authorized push payment liability gap. In the United States, consumers who voluntarily authorize a payment to a scammer have limited legal recourse. The UK's Payment Systems Regulator has moved toward mandatory reimbursement for APP fraud victims. The US has not. Until that policy gap closes, the migration Visa has documented will continue to produce losses for which consumers have no recovery mechanism.

How to protect yourself

The AuthentiLens editorial team has distilled the Visa Spring 2026 Threats Report, the executives' commentary, and this publication's broader research into six concrete protections for consumers navigating an AI-powered fraud landscape.

1. Treat Any Urgent Payment Request as a Scam by Default

The Visa report names urgency as the primary psychological tool of AI-enabled social engineering. Real banks, real merchants, and real federal agencies do not require you to act within ten minutes. If a message is pushing you to pay immediately, that urgency is the scam confirming itself.

• "Your account will be frozen in one hour" — scam.
• "Pay now to avoid late fees" — scam.
• "Your child needs bail money right away" — scam.
• "Limited-time offer, act now" — scam.

Real emergencies do not collapse if you take sixty seconds to verify. A scam will.

The specific warning signs that distinguish AI-generated urgent messages from legitimate ones are documented in our guide to recognizing fake bank, government, and family payment requests .

2. Verify Every Payment Instruction Through a Known, Separate Channel

A vendor email asking you to redirect a wire, a bank alert asking you to confirm credentials, a call from "your grandson" asking for bail money — all of these are vectors the Visa report identifies as primary AI-enabled scam delivery mechanisms.

The rule: do not call the number in the message. Do not reply to the email. Navigate to the organization's website by typing the URL directly into your browser, or call a number from your existing contacts. Verify through a separate, known channel before you send a single dollar.

For email-specific warning signs — including how to spot AI-generated phishing messages that no longer have grammatical errors or mismatched logos — see our guide on identifying AI-generated phishing attempts in your inbox .

3. Know Your Visa Zero-Liability Protection — and Its Limits

Most Visa cards extend zero-liability protection on unauthorized transactions when reported promptly. If you see a charge you did not make, call the number on the back of your card the same day.

The critical limit: zero-liability protection applies to unauthorized transactions — charges you did not approve. It does not apply to authorized push payment fraud, where you voluntarily sent money to a scammer. That is why the verification step before sending is essential. Once the authorized payment clears, the payment network has fulfilled its role. Recovery is between you and your bank, and outcomes vary.

4. Set a Family Verbal Password to Defeat AI Voice-Clone Scams

Fabara's "impersonations and scale" framing translates directly to the grandparent scam: an AI voice clone of your grandchild calls. The voice sounds exactly right. The distress sounds real. The request is for immediate wire transfer.

The defense is a family verbal password — a word or short phrase chosen in person, known only to immediate family, demanded on any emergency call before money moves. A real-time deepfake cannot know a password that was never spoken online or stored digitally.

Setting this up is especially important for older family members. Our complete guide on protecting elderly parents from AI-powered impersonation walks through the code-word system, the red-flag list to share, and what to do if a parent has already sent money.

5. Recognize AI Voice-Clone Calls in Real Time

The Visa report classifies vishing as a primary driver of the $1 billion scam total. AI voice cloning has made phone-based social engineering dramatically more convincing. Cloned voices no longer sound robotic. They breathe, pause, and mimic the exact cadence of the person being impersonated.

The defense is not tone-of-voice detection. The defense is procedural: any call asking for money or credentials triggers the verification protocol regardless of how real the voice sounds. Our guide on the warning signs that a phone call is a scam includes a step-by-step checklist for real-time call assessment.

6. Scan It With AuthentiLens Before You Send

You are not expected to become a forensic payment fraud analyst. That is what AuthentiLens is for.

When you receive an unfamiliar payment request, urgent email, vendor invoice, social-message demand, or account-locked alert:

• Paste the message text into the Scam Text Checker . We flag AI-generation signals, impersonation patterns, and urgency manipulation cues in seconds.
• Upload any attached image or document to the AI Image Detector . We detect AI-generated content, metadata inconsistencies, and fabrication signals — the same indicators the Visa report says platforms need to build into their own dispute workflows.

All of this runs in seconds, before you send money, before you share credentials, before the scammer receives a single dollar of the $1 billion Visa has measured.

What Visa, Banks, and Regulators Must Do

Consumer vigilance is necessary but not sufficient. The Visa report acknowledges the need for "continued network-level innovation paired with intelligence-driven defenses and coordinated action across the ecosystem." The AuthentiLens editorial team translates that corporate language into specific asks.

Close the authorized push payment liability gap. The single most impactful policy change available is mandatory reimbursement for APP fraud victims, as implemented in the UK. US consumers who are tricked into authorizing a payment currently have limited recourse. Changing that default would alter the risk calculus for both scammers and the platforms that could be doing more to stop them.

Integrate AI-detection tools into transaction monitoring. The same AI capability that generates convincing phishing emails can be deployed to detect them. Banks and payment networks should be scanning inbound messages, flagging urgency patterns, and warning customers before they send money — not processing the transaction and flagging it afterward.

Create a cross-platform fraud intelligence sharing system. A scammer phone number blocked by one bank is often active on six others. A fraudulent wire recipient account frozen by Visa rails is often already operating on competing networks. Sharing fraud intelligence across institutions in near-real-time would dramatically reduce the effective runway scammers have after their first successful attack.

Fund consumer awareness at the scale of the threat. The FTC's fraud reporting infrastructure and the FBI's IC3 complaint system are under-resourced relative to the $1-billion-per-six-months scale of the problem. Meaningful enforcement requires resources proportional to losses. The current gap between measured fraud and enforcement capacity benefits no one except the scammers.

The One Sentence Worth Keeping

Jabbara's "what once required deep technical skill can now be executed with a prompt" is the summary line for the entire 2026 fraud landscape. Every case this publication has covered this year — from the Centralia stock scam to the Lyft damage photo to the FIFA ticket clones to the AARP-documented grandparent calls — is a version of that one sentence played out in someone's life.

The prompt is free. The damage is not. The defenses exist. The question is whether consumers, platforms, banks, and regulators will deploy them at the scale the Visa data says is now required.