Detection catalog
The classes of problems Kaminari Ad detects — malvertising, scams, policy violations, and page behaviour.
Kaminari Ad groups what it finds into a few broad classes. Each finding attaches one or more tags to a scan.
Malvertising
Technical attacks delivered through an ad: drive-by downloads, forced auto-downloads, malicious redirects, cloaking (showing the scanner one page and users another), and known-bad domains flagged by domain-reputation and threat-intelligence sources.
Scams
Deceptive content aimed at users: fake giveaways and prizes, phishing landing pages, fake software updates, and impersonation of trusted brands.
Policy violations
Content that breaks your rules even if it isn't malware: prohibited verticals, misleading claims, missing disclosures, or anything your policy sets and custom rules define.
Occasionally the AI content analysis refuses to look at a landing at all:
its safety filter flags the page as prohibited content (almost always
explicit adult material) and returns no verdict. Such a scan gets the
Prohibited content (AI refused) tag (ai_prohibited_content, severity
medium) and finishes as partial, because it has no AI verdict or
classification.
Page behaviour
Some violations never appear on a screenshot — they are things the page does. Kaminari Ad watches the landing while it loads and tags:
- Back button hijack (
back_button) — the landing fills the browser history with copies of itself and listens for the Back button, so a visitor who presses Back is sent to another offer (in the same tab or a new one) instead of returning to where they came from. - Autoplaying sound (
autosound) — the landing tries to play sound as soon as it opens, without the visitor pressing play: autoplaying unmuted audio or video (often a synthetic voice), or audio started from a script. Browsers with a strict autoplay policy may mute it, but in-app browsers and webviews usually do not, so the attempt is what is tagged.
Other behaviour tags cover unsolicited pop-ups, forced redirects, automatic
downloads and native dialogs (alert, confirm, prompt).
Creative controls
Arrows, play buttons and close crosses on a banner can be real controls or
painted imitations, and a screenshot cannot tell which. Once the scan's own
click is done, Kaminari Ad clicks the creative's buttons and links found at the
points of a grid laid over it, once each. Every page load those clicks start is
stopped inside the browser until the check ends, so none of them opens the
landing. A creative that reports clicks with its own tracking requests may
still count them. It tags Interactive creative (creative_interactive,
severity low) when one of its buttons or links opens neither a tab nor a page
while the creative does open the offer elsewhere — through the scan's own click,
or, when that click opened nothing, through another of its links — for example
carousel arrows that switch slides while the slide itself opens the offer. A
banner that opens the landing wherever it is clicked, painted arrows included,
does not get the tag, and neither does one whose controls all open nothing.
The tag is informational. It is meant for custom rules:
a combo rule can drop a fake-interface verdict with none_of: [creative_interactive], so working carousel arrows are not reported as fake
UI.
Video (VAST) integrity
For VAST video ads, Kaminari Ad also flags problems specific to the format: an invalid or unparseable VAST document, a tag that only serves when fetched with a real browser (CORS-restricted / cloaked delivery), a wrapper chain deeper than the IAB limit, an embedded VPAID unit (which runs arbitrary JavaScript in the player), and video that fails to play. The clickthrough landing is scanned with the same malvertising, scam, and policy checks as any other creative.
A live, marketing-facing catalog is also published at kaminari.ad/detections.