Ad fraud is not a side issue in programmatic. It is a business, run by organized groups, with its own engineering teams and its own innovation cycle. Every time the industry closes a door, fraudsters look for the next one. In 2026, the next one is connected TV, and the newest weapon on both sides is AI.

The scale of the problem

DoubleVerify's 2026 Global Insights report, published in May, gives a clear picture of where things stand in CTV:

  • The number of CTV fraud schemes and variants grew 140% in the first quarter of 2026 compared with the same period in 2025.
  • More than 50 distinct bot attacks and variants were identified in 2025, and the number of fraudulent CTV apps detected was ten times higher than in 2024.
  • In campaigns with no protection, 9% of impressions were not served to a human. In protected campaigns, the figure was under 1%.
  • The report estimates that unprotected campaigns lose around 1.8 million dollars per billion CTV impressions to fraud.

At the same time, only 21% of surveyed advertisers said they use invalid traffic as a CTV KPI. Fraud is growing faster than the attention it gets.

Why CTV is such an attractive target

CTV combines everything a fraudster wants: high CPMs, large budgets and an environment that is hard to observe. A TV app does not run in a browser, so many of the signals that help verify a web impression are missing. A bid request can claim to come from a popular streaming app on a living-room TV, when in reality it was generated by a server in a data center.

Two cases show how far this can go.

Vastflux. HUMAN's Satori threat intelligence team uncovered and shut down an operation that spoofed more than 1,700 apps across 120 publishers and ran on nearly 11 million devices. At its peak, it generated 12 billion ad requests per day. The scheme injected code into video ads to stack dozens of invisible ads on top of each other, abusing the VAST video standard.

BADBOX 2.0. In March 2025, HUMAN, together with Google, Trend Micro and Shadowserver, partially disrupted a botnet of more than 1 million compromised consumer devices, mostly cheap, uncertified Android-based TV boxes, tablets and projectors. The backdoor came pre-installed on the devices, and among other things it loaded hidden ads and generated fake clicks.

The lesson from both cases: fraud now lives inside real devices and real supply chains, not only in obvious bot farms.

The toolbox, layer by layer

There is no single tool that stops fraud. The industry relies on layers, each catching what the previous one misses.

1. Authorization: ads.txt, app-ads.txt and sellers.json

The simplest tools are still among the most effective. With ads.txt and app-ads.txt, a publisher publicly lists which companies are allowed to sell its inventory. With sellers.json, each SSP and exchange lists the accounts it pays and whether they are publishers or intermediaries. With the SupplyChain object (schain), each bid request carries the list of companies it passed through.

Together, they let a buyer check that the app or site in a bid request really authorized the seller. They do not stop every scheme, but they make classic domain and app spoofing much harder, and they are the first thing a serious buyer checks.

2. Pre-bid verification

Verification companies such as HUMAN, DoubleVerify, Pixalate and IAS analyze traffic before a bid is placed. Their models look at device signals, IP addresses, data center ranges, app behavior and thousands of other patterns to decide whether a request is likely to come from a real person. Buyers use this as pre-bid segments, and supply platforms use it to filter traffic before it ever reaches a buyer.

The shift is important: instead of finding fraud in a report weeks later and asking for a refund, the bad request is never bought.

3. Takedowns and threat intelligence

Some of the biggest wins come from security research, not filtering. Teams like HUMAN's Satori map an operation, understand its infrastructure and then work with platforms and app stores to cut off its revenue. Vastflux and BADBOX 2.0 were stopped this way. These takedowns matter because they remove fraud at the source instead of blocking it one request at a time.

4. Accreditation and contracts

Third-party measurement accredited by the Media Rating Council gives buyers and sellers a common reference for what counts as invalid traffic. More and more contracts now include clear thresholds, refund rules and chargeback mechanisms based on these reports. This turns fraud from a debate into a process.

5. AI against AI

Fraudsters already use AI to generate fake sites and apps at scale. DoubleVerify's Fraud Lab reported a network of around 200 AI-generated websites designed to attract programmatic spend. The defense is also becoming AI-driven: models trained on billions of requests can spot patterns that no rule-based filter would catch, and they can adapt as quickly as the schemes do.

What this means for supply platforms

For an SSP, fraud prevention cannot be an option sold on top. It has to be built into how traffic is accepted and routed. At PeopleAds, every request is checked against ads.txt or app-ads.txt, sellers.json and the supply chain object, and our traffic is scanned by independent verification partners: Pixalate, DoubleVerify and HUMAN. Requests that fail these checks are not sent to buyers.

We also learned something simple: the traffic that converts is the traffic that can prove where it comes from. Clean, authorized supply is not only safer for buyers. It is also what buyers are willing to pay for.

Fraud will keep evolving. The goal is not to win once, but to make every new scheme more expensive to run than the last.

Sources