Hacking

Optimizely Injection: 178 Sites Served Malicious Code

Eng. Donya Bino Published  ·  11 min read

The trust model behind third-party experimentation tools is simple. You put a script tag on your page, the vendor serves the code, and you trust the vendor's certificate and infrastructure to keep that code safe.

That model held up for the certificate part. The snippets came from cdn.optimizely.com, over the vendor's own TLS. The problem was the contents.

Optimizely said early today that an attacker exploited a flaw in Web Experimentation to plant malicious code in scripts that customers' websites load from the service. Report URI, which monitors the scripts sites run, counted 178 organizations serving it to visitors yesterday.

The list includes CNN, the BBC, Zoom, Salesforce, GitLab, IKEA, HSBC, Equifax, Air France-KLM, Ledger, and Peacock. The vendor's own main site and login page also carried the code. Some researchers are calling the campaign OPTIMJACK.

How the Snippet Model Made This Possible

Sites that use Optimizely to run experiments load a script from the vendor called a snippet. Per the vendor's own documentation, the snippet updates itself whenever the customer changes a setting.

Each time it loads, it runs a field of custom code called Project JavaScript.

That is the injection point. Report URI found that the attacker wrote a loader into the Project JavaScript of customer projects and left the rest of each snippet untouched.

The snippet was still valid. It was still served from the vendor's CDN with the vendor's certificate. Every page that loaded one also ran the loader.

Some sites serve the snippet from their own domain rather than the vendor's. Those were affected too, because the content originates from the same project.

The Three Loader Variants

Report URI documented three versions, and the variation tells you the attacker anticipated blocking.

Variant

Mechanism

Most common

Loaded 1688360[.]cc/track.js directly

Base64 variant

Hid the address so a plain text search for the domain would miss it

Fallback

Carried a full copy of the payload to run if the domain was blocked

That third version is the reason a DNS block alone would not have stopped the campaign on every site.

What the Payload Did

The payload is a file called track.js, and it did three things.

Crypto address swapping. It replaced valid wallet addresses with the attacker's in page text, forms, links, the clipboard, and outgoing requests.

StepSecurity's analysis adds two constraints that matter. It swapped only an address that passed its chain's checksum and sat near payment wording like wallet, deposit, or send.

And copying was not safe. An address a visitor copied from one of these pages may already have been the attacker's by the time it was pasted into a wallet.

Card skimming. Added at 13:43 UTC, the skimmer captured card number, expiry date, and CVV when submitted through the site's own form fields, and sent them to the attacker's server.

It could not read cards entered in a cross-origin payment iframe, such as Stripe's hosted fields or Adyen's. That limit is worth knowing if you are trying to scope exposure.

Cookie and screenshot capture. The code sent the visitor's cookies and a page screenshot to the attacker, but only after it swapped an address near wallet or payment wording. It could not read cookies marked HttpOnly.

The iOS Exploit Delivery

On iPhones and iPads, the code loaded a second stage that Report URI identifies as an iOS exploit kit.

The mechanism: a script called boot.js loaded from two .icu domains. That script read the iOS version and opened an exploit chain in a hidden iframe.

iOS Version

Chain

13.0 – 17.2.1

Coruna

18.4 – 18.7

DarkSword

26.x

A separate chain that so far only collects device details

Between 17:13 and 17:57 UTC, the attacker also added a Coruna iframe to every visit from iOS 18.6 or earlier. Those pages reloaded every 60 seconds to retry the exploit, which is a behavior a visitor could notice.

The kit's own comments mark iOS 18.7.3 and later, and iOS 26.3 and later, as patched.

Cynet retrieved the second stage using an emulated iPhone browser. Its researchers describe what they saw as conditional delivery, explicitly not a successful compromise of a physical iPhone or iPad.

That distinction matters. Delivery and compromise are different claims, and only the first has been demonstrated.

The Money

Report URI says crypto has already been stolen. As of 19:25 UTC yesterday, the attacker's wallets had received about $7,040, all within roughly nine hours.

StepSecurity put the total across the same addresses at about $6,600.

Neither figure is a clean victim count, and StepSecurity is explicit about why. Two of the addresses were already in use before the injections began, and its analysts wrote that the chain does not show which transfers came from victims.

The Timeline

Time (UTC, Monday)

Event

04:21

First injection observed, in the BBC's snippet

13:43

Card skimmer added to the payload

17:13 – 17:57

Coruna iframe added for iOS 18.6 and earlier

~20:40

Mass cleanup begins

21:52

Only TalkTalk's snippet still carried the code

01:02 (today)

Optimizely publishes its incident notice

The first injection loaded webshell[.]lol, which Report URI calls the same actor's injector. That injector had also been planted through Kissmetrics, StorifyMe, and Minute Media scripts since late last week, and it first appeared on Turkish sites on September 17.

The Uncomfortable Detail

Smartsheet's script was cleaned once and then re-infected with a hidden version of the code.

That indicates the attacker still had access at that point. Report URI wrote its analysis before Optimizely's notice appeared, and said it did not know how the attacker got write access to so many unrelated customer projects. Its phrasing was direct. That is for Optimizely to explain.

What Optimizely Has Not Said

The incident notice confirms the flaw, the malicious JavaScript, the fix, and the removal. It also says the vendor secured affected user accounts and added protections against unauthorized code delivery.

It does not say:

  • How the flaw worked
  • How many projects were changed
  • When the fix went in
  • Whether the attacker still has any access
  • What customers should do

Those gaps are not cosmetic. The Smartsheet re-infection is the reason the fourth item matters, and customers cannot scope their own exposure without the second.

What Site Owners Should Check

Report URI's recommendations are specific.

  • Open Project JavaScript in every project, and custom code in every experiment. Look for 1688360, atob(, or any script you did not write. The atob( check catches the base64 variant.
  • Review who can write to your account. Users, API tokens, integrations. Then rotate their credentials.
  • Treat cards entered on Monday as possibly exposed if your own forms take card details.
  • Treat secrets in non-HttpOnly cookies as exposed, and rotate them.
  • Deploy a Content Security Policy with strict script-src and connect-src. Allowing cdn.optimizely.com does not help, because the code came from that host. strict-dynamic or unsafe-eval let the payload run on some sites.
  • Consider telling your customers that your site served them the code.

StepSecurity adds an important framing. A snippet that is clean now may still have served the code on Monday.

  • Block the indicators below and search logs back to 08:00 UTC Monday.
  • Fetch every snippet you serve, search for 1688360, .icu, and track.js, and record its Last-Modified header.
  • Ask Optimizely which of your project IDs were changed, and when and how.
  • Keep third-party tags off checkout pages, and take card details through the payment provider's own cross-origin iframe.
  • Replace any card entered on an affected site Monday, and verify wallet addresses on a hardware wallet's screen rather than against copied text.
  • Consider having iOS 18.6 or earlier devices checked for signs of compromise if they visited affected sites.

What Visitors Should Do

  • Update iPhones and iPads to iOS 18.7.3 or later, or iOS 26.3 or later.
  • Compare any crypto payment made Monday against the address you intended to pay. If they differ, the address may have been swapped.
  • Watch card statements if you paid by card on one of the affected sites.

The Indicators

Type

Value

Domain

1688360[.]cc — payload host, with paths /track.js, /card.php, /hit.php

IP

136.113.106[.]136 — address 1688360[.]cc resolved to

Domain

5vobauua3ojpklx[.]icu — iOS stage

Domain

eil786npto9u5bq[.]icu — iOS stage

Domain

webshell[.]lol — injector loaded by the first injection at 04:21 UTC

SHA-256

f7b84b9a94f749f8ced3f0112d0a66894e8bbc3b4aa5592021d1b8976a9bad40 — final track.js

SHA-256

617df364120adc47fb4f55dcc0c69a9b6c9e108a92c60bc2b64a266254ab02f3 — earliest captured track.js

Two cautions on using these.

The payload changed multiple times during the day, so a single file hash will not capture every version.

And Cynet treats cdn.optimizely.com as a legitimate delivery host rather than an indicator. It advises against broad blocks of that host or of the affected news sites. Blocking the CDN would break every customer still using the service legitimately, and it would not have stopped the base64 or fallback variants anyway.

The Structural Lesson

A tag manager or experimentation platform is a write-access channel into every page that trusts it. That is the product. It has to be able to change what your site runs, on demand, without a deploy.

The security control that matters is not the certificate. It is who holds write access to the project, and what happens when that access is used against you.

Report URI's list of 178 organizations is a measure of how many sites made the same reasonable decision to trust a vendor's CDN.

FAQ

What happened?

An attacker exploited a flaw in Optimizely Web Experimentation to inject malicious JavaScript into the Project JavaScript field of customer projects. Snippets served from the vendor's CDN then ran that code on visitors' browsers.

How many sites were affected?

Report URI checked 1,611 snippets and found code in 188, belonging to 178 organizations. It says the real number is likely higher, because it could only check the snippets it found in time.

Which organizations were listed?

CNN, the BBC, Zoom, Salesforce, GitLab, IKEA, HSBC, Equifax, Air France-KLM, Ledger, and Peacock among others, plus Optimizely's own main site and login page.

What did the code do?

It swapped crypto wallet addresses near payment wording, skimmed card details from the site's own form fields, captured cookies and page screenshots, and on iOS loaded an exploit chain.

Did the card skimmer work everywhere?

No. It could not read cards entered in a cross-origin payment iframe, such as Stripe's hosted fields or Adyen's.

Were phones actually compromised?

Cynet describes what it observed as conditional delivery, explicitly not a successful compromise of a physical iPhone or iPad.

How much crypto was stolen?

Report URI reported about $7,040 across the attacker's wallets as of 19:25 UTC. StepSecurity put the total at about $6,600 and noted that two of the addresses were already in use before the injections began, so the chain does not show which transfers came from victims.

Why did one site get re-infected?

Smartsheet's script was cleaned and then re-infected with a hidden version. Report URI says that indicates the attacker still had access at that point.

Does the vendor's notice explain the flaw?

No. It confirms the flaw, the fix, and the removal, but does not say how the flaw worked, how many projects changed, when the fix went in, whether the attacker still has access, or what customers should do.

Is the snippet still dangerous if it looks clean now?

Possibly. StepSecurity notes that a snippet that is clean now may still have served the code on Monday, which is why log review back to 08:00 UTC matters.

Should I block cdn.optimizely.com?

Cynet advises against it. That host is the legitimate delivery path, blocking it breaks the service for everyone still using it, and it would not have stopped the base64 or fallback loader variants.

What is the single most useful check?

Open Project JavaScript in every project and search for 1688360, atob(, or any script you did not write.

What about cookies?

The code could not read HttpOnly cookies. Anything kept in a cookie without that flag should be treated as exposed and rotated.

Why does CSP not fully help?

Because the code came from cdn.optimizely.com, which you presumably allow. strict-dynamic or unsafe-eval would let the payload run on some sites regardless.

Source: The Hacker News
Professional Services

Explore Our Cybersecurity Services

Our insights are backed by hands-on service delivery. If your business needs professional cybersecurity support, our UK-based specialists are ready to help.

© 2016 – 2026 Red Secure Tech Ltd. Registered in England and Wales — Company No: 15581067