deniz.in

Markets

Weather

Loading weather

· via Hacker News – Front Page (native)

Skimmer injected into Optimizely bundles hit visitors of CNN, BBC and Zoom

Injected code in Optimizely experiment bundles served a crypto clipper and card skimmer to visitors of CNN, BBC, Zoom and a dozen other major sites, a researcher's analysis shows.

Skimmer injected into Optimizely bundles hit visitors of CNN, BBC and Zoom

Malicious JavaScript planted inside Optimizely experiment bundles served a combined crypto clipper and card skimmer to visitors of at least 15 high-traffic websites, including CNN, the BBC, Zoom, Salesforce and GitLab. According to a technical write-up published on amirulcyber.com, which reached the Hacker News front page on 16 November 2026, the campaign put millions of exposed visitors in reach of a payload its author calls OPTIMJACK.

How the injection works

Optimizely supplies A/B testing and experimentation code that sites embed as a third-party script. The researcher found that inside affected bundles, directly after the legitimate experiment code, a short loader creates a new script element pointing to a track.js file on the domain 1688360.cc and appends it to the page. The same injected line appeared in GitLab's Optimizely project and in ten further project or snippet IDs belonging to unrelated organisations, pointing to a compromise at the tag layer rather than at any single customer's servers. The write-up does not establish how the loader got into the bundles in the first place.

What the payload does

The recovered track.js is one self-contained script with four modules, per the analysis:

  • A clipper that validates cryptocurrency addresses with genuine checksum routines (including double SHA-256, base58check and bech32) covering ETH, BTC, BCH, LTC, TRX, DOGE, SOL and TON, then substitutes attacker-controlled wallets. It rewrites visible page text on a 50-millisecond sweep gated by payment keywords, input and textarea values including framework-managed state, link and image URLs, copy events, the clipboard API, navigator.share, and the URLs and bodies of fetch and XMLHttpRequest calls.
  • An exfiltration module that posts the current page URL (capped at 500 characters), cookies (capped at 8,000), a JPEG screenshot rendered with an html2canvas copy fetched from the same command-and-control host, and every original wallet address the clipper replaced, to an endpoint at /hit.php.
  • A card skimmer that applies Luhn and card-brand checks plus expiry and CVV format validation, intercepts fetch and XHR POST bodies (JSON, FormData and nested objects) and form submissions, deduplicates per session, and sends results via sendBeacon to /card.php.
  • An iOS second stage: on iPhone, iPad and iPod only, the script pulls an obfuscated boot.js from two .icu hosts and, on iOS 18.6 or older, adds a hidden iframe. The researcher says boot.js is still being analysed.

Who was affected

The write-up documents 15 captured page loads: CNN, Zoom, the BBC, Salesforce, GitLab, American Eagle, Hootsuite, Smartsheet, Air France's Italian site, Virgin Media O2, Stamps.com, Tricentis, Mutual of Omaha, HSBC's Singapore insurance portal and Fox News. Twelve of the fifteen were served through cdn.optimizely.com bundles; the exceptions were Fox News, reached via an advertising wrapper on cdn.confiant-integrations.net, American Eagle via a first-party bundle on its own domain, and Air France via a Jscrambler-protected script. Two of the affected sites escaped execution because their Content-Security-Policy blocked the unknown host, producing script-src violations and failed requests instead of a running payload.

What operators should check

The researcher proposes a three-step audit: fetch the Optimizely bundle actually served to visitors and search it for the string 1688360; alert on CSP script-src violations that name unknown hosts; and paste a canary wallet address on a staging page to confirm the submitted value matches what was typed. The write-up also publishes indicators of compromise, including the .cc and .icu hosts, the endpoint paths, and SHA-256 hashes for track.js and one boot.js sample.

Why it matters

The defining property of this attack is that a site's own infrastructure can be entirely clean while its visitors are robbed: the malicious code runs inside a third-party tag with full page privileges. Consent tooling is no defence here — the researcher observed that the OneTrust consent layer present on the page classified the GitLab script into a category without ever inspecting it. Content-Security-Policy was the one control that made a difference, and only two of the fifteen sites had it working in their favour. For any organisation shipping third-party marketing scripts, the incident is a concrete argument for treating the tag layer as part of the attack surface, monitoring script-src violations closely, and never assuming that a clean codebase equals safe visitors.

  • #supply-chain-security
  • #web-security
  • #malware
  • #optimizely
  • #javascript