Optimizely malware: how to protect your site from third-party scripts

Your testing tool loads on every page.

It runs before your checkout.

On 16 November, for 178 organisations, it ran someone else's code.

That isn't a reason to panic or rip out your experimentation stack. It is a reason to treat every script on your site as something that can change without you touching it.

TL;DR:

  • What happened: An attacker added malicious JavaScript to the Project JavaScript of Optimizely Web customer projects. It was served from Optimizely's own CDN for most of 16 November [1][2].
  • What it did: Swapped crypto wallet addresses, skimmed card details from site forms, stole non-HttpOnly cookies, and served iOS exploit kits [1].
  • Who's exposed: 188 snippets across 178 organisations were found infected. The true number is likely higher [1].
  • What to do: Audit your snippet, lock down tool access, rotate exposed secrets, and deploy a Content Security Policy that restricts where data can be sent [1].

What happened in the Optimizely malware incident?

An unauthorized party exploited a vulnerability in Optimizely's Experimentation product and added malicious JavaScript to some customer projects [2]. Optimizely says it has fixed the vulnerability, removed the injected code, secured affected accounts, and notified affected customers [2].

Report URI's Scott Helme reconstructed the timeline from crawl data [1]:

  • 04:21 UTC: The first injection appeared in a customer snippet.
  • 13:43 UTC: A card skimmer was added to the payload.
  • 17:13 UTC: An iOS exploit kit was added.
  • 19:24–19:59 UTC: A sweep of 1,611 Optimizely snippets found 186 infected.
  • 23:11 UTC: All infected files were reported clean.

The payload was rebuilt 26 times in one day [1]. Named brands in the report include Zoom, BBC, IKEA, HSBC and Salesforce, and Optimizely's own sites were infected too [1].

This wasn't CDN cache poisoning. The rest of each snippet was untouched, and the files were served from Optimizely's CDN with a valid certificate [1]. To every browser and most security tools, the code looked legitimate.

What did the malicious code do?

It ran four kinds of attack at once [1]:

  1. Crypto address swapping: Replaced real wallet addresses in page text, forms, the clipboard and outgoing requests with the attacker's.
  2. Card skimming: Captured card number, expiry and CVV from forms on the site and sent them to the attacker. Hosted payment iframes were not affected.
  3. Cookie and screenshot theft: Sent page URLs, non-HttpOnly cookies and page screenshots to the attacker's server.
  4. iOS exploit kits: Served device exploits to older iOS versions. Report URI says iOS 18.7.3+ and 26.3+ are patched.

On-chain data shows about $7,040 received by the attacker's wallets on the first day [1]. That figure says nothing about how many cards, cookies or devices were compromised [1].

Is your site affected?

If you ran Optimizely Web on 16 November 2026, check now rather than wait for an email. Optimizely says affected customers have been notified [2], and it posts updates on its status page [3].

Check these first [1]:

  • Your Project JavaScript and experiment code: Look for 1688360.cc, atob( and any script you don't recognise.
  • Your logs and network data: Search for the domains 1688360.cc, webshell.lol and linkforge.cc, and block them, including subdomains.
  • Your checkout: If customers typed card details into a form on your own page that day, treat those cards as potentially exposed.

The same actor used other vendors' scripts in an earlier wave on 13–16 November, including Kissmetrics, StorifyMe and Minute Media [1]. Not an Optimizely customer? Check your other third-party scripts anyway.

Why didn't most security headers stop it?

Because the code came from a host every site had already trusted. Report URI checked the hosts of 256 affected organisations, and only 15% blocked the payload [1].

CSP result

Organisations

What it means

No CSP at all

68

Nothing stopped the payload or the data theft [1]

CSP allowed payload and data theft

50

The policy existed, but trusted the wrong things [1]

Payload ran, but connect-src blocked data theft

17

Code ran, data couldn't leave [1]

Payload blocked

24

The policy worked [1]

Allowing cdn.optimizely.com didn't help, because the malicious code was served from that trusted host [1]. Report URI calls out 'unsafe-eval' and 'strict-dynamic' as weaknesses. In one case, an inline fallback ran through eval even though the main script was blocked [1].

The lesson: a script allowlist alone can't tell good code from bad code on the same domain. Restricting where data can be sent (connect-src) is what saved the 17 organisations in the middle row.

Subresource Integrity (SRI) won't solve this on its own either. SRI pins a file to a known hash [5], but testing snippets change every time you publish an experiment, so a fixed hash breaks with each legitimate change. That's our reading, not a finding from the report.

How do you protect your website from third-party scripts?

Treat every script as a dependency that can change without you. These seven steps apply to any client-side tool: Optimizely, VWO, AB Tasty, Convert, tag managers, chat widgets and review apps.

1. Inventory every script on your site

List each third-party script, who owns it internally, and which pages it loads on. PCI DSS v4.0 already requires an inventory and authorization of every script on payment pages, plus change detection [6]. Most brands that take cards should be doing this already. If you'd like an outside view of what runs on your funnel pages and where it adds friction, start with a CRO audit.

2. Keep testing scripts off payment pages

The skimmer read card data from forms on the site, but hosted payment iframes were not affected [1]. Use your payment provider's hosted fields or hosted checkout. Exclude checkout and account pages from your testing snippet unless you're actively testing there.

3. Lock down access to your testing tool

The attacker got write access to customer projects [1][2]. Report URI recommends auditing users, API tokens and integrations, then rotating credentials [1]. Remove ex-staff and old agencies, enforce SSO or two-factor authentication, and give publish rights to as few people as possible.

4. Deploy a Content Security Policy that controls where data goes

Start in Report-Only mode so nothing breaks, then tighten [4]. Prioritise a restrictive connect-src, avoid 'unsafe-eval' where your tools allow it, and turn on violation reporting [1]. If a vendor's documentation requires 'unsafe-eval', ask them why, and weigh it against the risk.

5. Monitor scripts for change

Report URI notes that a CSP in Report-Only mode is enough to flag new script sources, new data destinations, new frames and changed script fingerprints [1]. Whichever tool you choose, someone should get an alert when the code running on your checkout changes.

6. Protect cookies and secrets

The malware stole cookies readable by JavaScript [1]. Set session and auth cookies as HttpOnly and Secure. Never keep API keys or tokens in cookies or front-end code. If anything sensitive sat in a non-HttpOnly cookie on 16 November, rotate it [1].

7. Have a response plan before you need it

Decide now who can remove a script from the site in minutes, not hours. Report URI also suggests considering customer notification if you were affected [1]. Check your obligations with your legal or compliance team.

Pro tip: run a crawl of your own checkout today and compare the scripts it loads against the list your team thinks it loads. The gap is usually the first finding.

Should you stop A/B testing?

No. Stopping testing doesn't remove the risk. It moves it to your chat widget, your review app or your tag manager. Every client-side script carries the same exposure, and the earlier wave of this attack used three non-testing vendors [1].

What changes is how you run testing:

  • Server-side, edge or native testing keeps experiment logic out of third-party browser scripts on your highest-risk pages. On Shopify, native Rollouts testing is one option.
  • Tool choice should include security: SSO, role-based permissions, audit logs, and change history. Compare tools in our A/B testing software directory.
  • Governance matters as much as the tool. Fewer publishers, reviewed code, and a clean list of what's live. If your team runs tests in-house, build these rules into how it works; see CRO training.

If you're weighing whether to run testing in-house or with a managed CRO partner, the same question applies to both: who reviews the code that ships, and who notices when it changes?

Summary: the tool wasn't the only failure. Broad access, permissive policies and no change alerts let it run for hours.

Where to start this week

Pull the list of scripts on your checkout, review who can publish in each tool, and put a Report-Only CSP live on your payment pages. Then check the iOS version on your own devices [1].

If you'd like a second pair of eyes, Wavesy offers a free 15-minute Loom review of your website [7]. We'll flag the scripts running on your key funnel pages along with the conversion friction. If you want a testing program built with governance from day one, see our Conversion Rate Optimization service [8]. If you first need a ranked plan of what to test and in what order, our CRO Strategy maps it out.

Once your scripts are under control, the next question is what to test. Read which changes to test first to improve your DTC conversion rate.

One script list and one access review is enough to close the door this attack walked through.

Sources

PUT THE THINKING TO WORK

Your next experiment
deserves a clearer plan.

Turn your customer evidence into a prioritized CRO roadmap, with hypotheses, ownership, and a useful next step.

Explore CRO Strategy
Filip Doneski, Partner and Lead CRO Strategist at Wavesy
ABOUT THE AUTHOR

Filip Doneski

Partner & Lead CRO Strategist at Wavesy

Senior CRO Strategist specializing in A/B testing, data-driven growth, psychology & conversion optimization for DTC, Retail and SaaS.

More from this author