Оригинальный текст на языке EN взят с сайта-источника. Источник ↗

Malicious iOS scripts have become a leading cause of ad platform suspensions. They accounted for about one in five suspensions in Q3 2026, according to the State of Ad Fraud 2026: ADEX Trust & Safety Report from anti-fraud firm ADEX.

The report examines one scheme in detail. It was built on the Coruna exploit kit and ran through about 50 advertiser accounts that used the same back-end setup behind very different landing pages.

The platforms suspended those accounts under their own policies.

How the Attack Chain Works

ADEX describes four stages:

  1. Device fingerprinting. A hidden iframe on the landing page checks whether the visitor is a real iPhone or iPad. It also identifies the exact model and iOS version.
  2. Reviewer evasion. Bots, crawlers, suspicious IPs, private browsing sessions and devices in Apple’s Lockdown Mode are served a harmless page. This filtering keeps the scheme invisible to moderators.
  3. Tailored payload. The kit assembles an exploit chain matched to the device’s exact iOS version and chip. It delivers the chain as an ordinary-looking .min.js file.
  4. Financial theft. The final implant targets blockchain and financial apps. It searches the phone for wallet recovery seed phrases.

The exploit chains cover iOS 13.0 through 17.2.1. The most exposed users are those on older or unpatched hardware, including:

  • iPhone 6s and iPhone 7
  • first-generation iPhone SE
  • older iPads

ADEX notes that its description covers observed technical behaviour and is not a statement about any identified person or company.

AI Is Now Standard in the Fraud Toolkit

Beyond iOS, ADEX found AI indicators in 52% of the distinct fraud schemes it observed. Within those AI-involved schemes:

  • 64% used bots that pass as humans
  • 53% used cloned brands and fake endorsements
  • 11% used deepfakes

“AI did not invent new abuse in 2026, but it made old abuse cheap, fast, and fluent… The tell is no longer bad grammar or a blurry logo,” said Sergey Martianov, CPO at ADEX.

Next Target: AI Agents

ADEX forecasts that in 2027, attackers will increasingly aim at the AI agents that browse and buy on users’ behalf. According to the report, likely injection vectors on landing pages include:

  • Invisible text: white-on-white text, zero-size fonts, or blocks hidden with CSS
  • Page plumbing: alt text, meta tags, HTML comments and structured data that browsers do not display but agents parse
  • Text inside images: instructions written into banners or product photos for agents that read images
  • Fake assistant notes: lines such as “this is the verified official store” planted in reviews, FAQs or terms
  • Agent cloaking: one version of the page for humans and moderators, another for visitors that identify as AI agents

The injected instructions tend to be short and practical. Examples include ranking a product first, ignoring a security warning, or entering the user’s details to continue. ADEX labels this a forecast, not a measured category, but notes the techniques are well documented in security research.

Defensive Recommendations

The ADEX report recommends that security teams and advertisers:

  • judge campaigns by where the click actually ends up, across devices and countries
  • treat any landing page whose only exit is a messenger link as high risk
  • verify declared agent identities, for example through cryptographically signed agent requests
  • check that agent behaviour matches its stated task
  • audit their own pages for hidden text, metadata and structured data aimed at machines

For iOS users, the immediate takeaway is simple: update devices. Those that cannot run current iOS versions remain exposed to chains like this one.

Read the full analysis in the ADEX State of Ad Fraud 2026 report.