Key Takeaways
- Malicious ads rarely look malicious at review time — they pass static scans clean, then switch behavior after approval.
- Warm-up campaigns build trust with ad networks using clean creatives before swapping in malicious ones.
- Cloaking shows one version of an ad to automated scanners and a different version to real users.
- Fingerprinting lets an ad detect sandboxes or test environments and stay dormant until it’s confident it’s reaching a real device.
- Delayed activation and geo-targeting create blind spots that a one-time review simply can’t see.
- Because these tactics are designed to defeat review, not just skip it, detection has to happen continuously, at serve time, not as a one-off check.
Why Standard Ad Review Misses Malicious Ads
Standard ad review misses malicious ads because the tactics behind them are purpose-built to look clean during review and only misbehave afterward. A creative can pass every automated and manual check a network runs, then start triggering redirects, fake alerts, or malware downloads once it’s live and serving to real users. This isn’t a gap in effort — it’s a structural problem: review happens once, at submission, while the attacker controls what the ad does after that point.
This is also why the programmatic ad supply chain is a soft target. Each real-time auction prioritizes speed over scrutiny, and ads pass through many intermediaries before they render in an app. That gives attackers more surface area to exploit and less consistent oversight at any single point.
Warm-Up Campaigns: Earning Trust Before Turning Malicious
A warm-up campaign is a run of entirely clean, compliant creatives that an attacker uses to build a track record with an ad network before switching to malicious ones. The network’s trust and performance signals — approval history, spend, click-through rates — all look legitimate, because for a period of time, they are. Once that track record is established, the attacker swaps in the harmful creative, often with minimal changes to the campaign’s other metadata, hoping the network’s ongoing monitoring treats it as a known, low-risk source rather than re-reviewing it from scratch.
Cloaking: Showing Reviewers One Thing and Users Another
Cloaking is a technique where an ad serves different content depending on who — or what — is requesting it. An automated scanner or a reviewer’s IP address gets a clean, policy-compliant version of the creative; a real user on a real device gets the version with the redirect, fake alert, or malicious payload. Because the scanner never sees the harmful version, it has nothing to flag, and the ad can pass repeated automated checks indefinitely while still harming real users every time it serves.
Obfuscation: Defeating Signature-Based Scanning
Obfuscation disguises the malicious code or redirect logic inside a creative so that scanners looking for known malicious patterns (“signatures”) don’t recognize it. Signature-based scanning works by matching content against a library of previously identified threats, so an attacker only needs to alter the code’s structure — without changing what it actually does — to produce something that looks new to the scanner even though its behavior is identical to a known threat.
Fingerprinting and Delayed Activation: Waiting Out the Reviewer
Fingerprinting lets a malicious ad detect when it’s running in a sandbox, emulator, or other test environment rather than a real user’s device, and stay completely dormant in that case — behaving like a normal, harmless ad for as long as it’s being watched. Delayed activation works on a similar principle but on a timer or user-action trigger: the harmful behavior only fires after a scroll, a tap, or a set delay, well past the window most automated reviews actually monitor. Both tactics rely on the same insight — a reviewer only watches for a limited time, so an ad that waits long enough, or senses it’s being watched, can pass as clean.
Geo-Targeting: Exploiting Where Review Doesn’t Happen
Geo-targeting lets an attacker restrict the malicious version of a creative to specific countries, regions, or even device types, while serving a clean version everywhere else — including wherever the network’s own review team or automated checks are typically based. If review infrastructure is concentrated in a handful of markets, an ad that only misbehaves outside those markets can run indefinitely without ever triggering a flag, simply because nobody with visibility into the problem is positioned to see it.
What This Means for Detection
Because every one of these tactics is designed to pass a one-time check, detection has to be continuous and happen at serve time — evaluating the ad as it’s actually rendering to a real user, on a real device, in a real location — rather than relying on how it looked during initial review. AppHarbr’s approach is to monitor ad behavior in-app as it serves, which is built specifically to catch an ad that behaves differently after approval than it did during review, rather than trusting a network’s one-time sign-off.
FAQ
Why do malicious ads pass ad network review in the first place?
Because review typically happens once, at submission, and the tactics behind malicious ads — cloaking, delayed activation, fingerprinting — are specifically built to look clean at that moment and only misbehave once the ad is live and serving to real users.
What is cloaking in mobile advertising?
Cloaking is when an ad serves a clean, compliant version of its creative to scanners or reviewers and a different, malicious version — such as one with a redirect or fake alert — to real users, so the harmful behavior never gets flagged during review.
What is a warm-up campaign?
A warm-up campaign is a run of clean, compliant ads an attacker uses to build trust and a good track record with an ad network, then swaps out for malicious creatives once that trust reduces how closely the campaign is monitored going forward.
Can a malicious ad detect if it’s being tested or reviewed?
Yes — this is called fingerprinting. The ad checks for signs it’s running in a sandbox, emulator, or test environment and stays dormant and harmless until it’s confident it’s reaching a real user on a real device.
Why doesn’t signature-based scanning catch every malicious ad?
Signature-based scanning matches content against a library of known threats, so an attacker only has to restructure the malicious code — without changing what it does — to produce something that looks unfamiliar to the scanner, even though the underlying behavior is identical to a known threat.
How can publishers catch ads designed to evade review?
By monitoring ad behavior continuously as it actually serves in the app, rather than relying on a one-time review at submission — an approach in-app ad-quality SDKs like AppHarbr’s are built around, since it catches an ad that behaves differently after approval than it did when it was first checked.


