The Maze: Optimizely says a signup flaw let an attacker impersonate customer administrators and place malicious code in its Web Experimentation service on November 16. Businesses use the service to test different website versions. Its code runs in visitors’ browsers, so an altered experiment could reach shoppers without a change to the merchant’s own website files. The company says it fixed the flaw. Report URI found infected snippets belonging to 178 organizations; that is not a count of retailers or stolen payment cards.

Source: Optimizely. Documentation example of the History view, with change filters and a code comparison; this is not an incident log.
The details
🧪 A testing account became a route into the page. Optimizely’s initial incident report says the attacker created accounts, then exploited the signup process to act as existing administrator users without their passwords. It says the attacker guessed variations of employee email addresses, repeating attempts across customers. Inside affected projects, the attacker changed Project JavaScript: custom code delivered alongside the site’s experiments. Report URI’s analysis explains why the delivery address offered false comfort. The modified code still arrived from a trusted Optimizely host. The company says Web Experimentation was affected; Feature Experimentation, its software development kits and legacy Full Stack were not.
💳 Where shoppers typed their card details mattered. Report URI found that the malicious code could collect card numbers, expiry dates and security codes entered into the site’s own form fields. Card fields inside hosted payment frames, such as those supplied by Stripe or Adyen, were outside its reach. These frames isolate the payment provider’s form from other code on the page. The payload also replaced cryptocurrency payment addresses and targeted older iPhones and iPads. The researcher found 188 infected files in 1,611 discoverable snippets, not a representative sample of all customers. Its list included ASOS and American Eagle, but neither a listed brand nor a malicious script proves that cards were stolen.
🔎 Check customer records before declaring an exposure window. The company report offers customer-specific timelines through account managers and recommends preserving local logs. Its claimed first malicious write, at 07:51 UTC, is later than Report URI’s earliest observed injection, at 04:21 UTC. Their cleanup timestamps and published suspicious-domain lists also differ. Neither gives a reconciled window for every site. Optimizely’s History view records Project JavaScript changes and filters activity by date, type and whether it came through the interface or an API. The company now says password and API-token resets are unnecessary and that re-enabling the service is safe. Those are updated company assessments during an ongoing investigation.
Our take
Website experiments deserve the same scrutiny as production code. A testing tool can change what a shopper’s browser does. The useful decision is how much authority it needs on each page, and which safeguards still work if the vendor’s code turns malicious.
🚧 Test execution and data exit separately. In Report URI’s policy assessment, 17 organizations blocked the script’s data transmission while still letting it change wallet addresses inside the page. Another 24 blocked the payload altogether. A Content Security Policy is a set of browser rules controlling permitted code and destinations. A restrictive destination list can contain some damage; it cannot undo a fraudulent address already displayed to a customer. Security teams should test those two outcomes separately. These are findings from the researcher’s assessed sites, not a measured protection rate for all merchants.
✂️ Reduce access where the experiment adds no value. Optimizely recommends removing its snippet from sensitive-input pages where no experiments are running. That gives checkout teams a concrete review: does this page need the testing code at all? The report also describes approval before publishing Project JavaScript as a feature request. Do not build a control plan around an approval feature presumed to exist. An internal review process and narrower page coverage are proposals operators can assess now, rather than proof that any remaining integration is safe.
📁 Keep evidence after the vendor cleans up. Salesforce told BleepingComputer it was disabling scripts across preview and production while investigating. That precaution did not establish stolen customer data. Equally, Optimizely’s cleanup does not establish that nothing was taken: the company says it can see code loads but cannot determine what the code captured. Retaining site, payment and browser-policy logs therefore has a different job from removing the bad script. One stops delivery; the other helps establish what happened to visitors.


