You Gave Your Marketing Vendor a Pen That Writes Code Into Your Customers’ Browsers

Nov 18, 2026 | Transhuman / Security

The Optimizely hijack was not a JavaScript problem. It was a write-access problem that nobody on the security team ever counted as privilege.

When I was on the other side of the fence, getting my own JavaScript onto a big website was the whole game. You hunted for a stored cross-site scripting bug, a sloppy comment field, an admin panel somebody forgot to lock down. You worked for it, because once your script ran in that origin, every visitor’s browser treated you like the site itself. That was the prize. Cookies, form fields, whatever the page could touch, you could touch.

On November 16, somebody skipped all of that work. They did not break into the BBC, CNN or Zoom. They got write access to a field inside an A/B testing product, and those sites loaded the result for them, signed with a legitimate certificate, from a domain everybody had already approved [1][3].

I want to talk about why that happened, because the industry conversation is already sliding toward the wrong lesson.

What actually happened

Here are the facts as the researchers and the vendor have described them. Report URI’s Scott Helme found malicious JavaScript served to visitors of at least 178 organizations through Optimizely’s Web Experimentation platform [1]. His team checked 1,611 Optimizely snippets and found the code in 188 of them, and they say the real number is probably higher because they could only check what they found in time [2][4]. The list reads like a who’s who of the web: Salesforce, GitLab, IKEA, HSBC, Equifax, Air France-KLM, Ledger [3][4].

Optimizely’s own statement says an unauthorized party “exploited a vulnerability in the Optimizely Experimentation to add malicious JavaScript to some customer projects,” and that it blocked the vulnerable path, deployed a fix, removed the injected code, ended active sessions, revoked tokens and reset access for affected accounts [1]. The Register reports that the company tied the hole to the sign-up process for its experimentation software, a flaw researchers are calling OPTIMJACK [2]. The vendor has not published a full root cause, and its incident notice did not say how the flaw worked or what customers should do [3]. Keep that gap in mind. Everything below is built on what is confirmed, not on guesses about the bug.

The mechanics matter. Sites using the product load a snippet from the vendor, and every time it loads it runs a field called Project JavaScript [3]. The attacker wrote a loader into that field and left the rest of the snippet alone, so the code kept coming from cdn.optimizely.com with the vendor’s own certificate [3]. Sites that hosted the snippet on their own domain were hit too, because the content came from the same project [3].

Then the payload grew up over the course of a single day. Helme says it was rebuilt 26 times [1][4]. It swapped cryptocurrency wallet addresses for the attacker’s, a card skimmer showed up in the afternoon, and it grabbed cookies and screenshots [1][3]. On older iPhones and iPads it tried to load the Coruna and DarkSword exploit kits [1][2]. To be precise about that last part, the researchers describe delivery of those kits, not confirmed takeover of real devices [3][4].

The cash haul was small. Estimates of what landed in the attacker’s wallets run from roughly $6,600 to about $7,150 depending on who did the counting, and nobody can see how many cards were skimmed [3][4]. Do not let the dollar figure fool you. A few grand in crypto is what a test run looks like.

Everybody is about to say “deploy CSP”

The loudest takeaway is Content Security Policy, and I get it. Zach Edwards told Krebs that most sites outside the biggest e-commerce, social and financial platforms do not bother with CSP or leave it wide open for marketing reasons [1]. Helme’s numbers back him up: of the affected organizations, 68 had no CSP at all, 50 had a policy that still let the payload run and phone home, and only 24, about 15 percent, had a policy strict enough to block it [1][4].

So yes, CSP. But read the next part twice, because it is the whole story.

Allowing Optimizely’s domain in your policy offered no protection, because the malware came from that domain [4]. Report URI’s own guidance says the same thing: allowlisting cdn.optimizely.com does not help, and settings like strict-dynamic or unsafe-eval let the payload execute on some sites [3]. Help Net Security notes that Optimizely’s documentation recommends a setting that permits eval, and that setting let a backup copy of the malicious code run on HSBC’s site [4].

A CSP that trusts a hostname is an access control list with one entry, and that entry says “whoever controls this vendor project.” It does not say who that is. It cannot. The sites that survived were the ones whose policy also constrained where the code could send data and what it could load next, not the ones that merely trusted the right host [3].

The real privilege was the pen, not the domain

Here is the part I want every security leader to sit with. Whoever can write to that Project JavaScript field can execute arbitrary code, as your origin, in every customer browser that loads your page. That is not a marketing setting. That is one of the most powerful privileges your company owns. It is closer to production deploy rights than to a dashboard login.

Now ask how it was granted. Somebody in growth or digital marketing signed up for a tool. They got seats. Maybe an agency got seats. Maybe an integration got an API token so experiments could be pushed automatically. Edwards points out that this class of tool is built to fire immediately on page load, before any consent dialog, so it renders first [1]. The product is designed to run early, everywhere, with no friction. That design is the business value, and it is exactly what an attacker wants.

Did any of those seats go through a privileged access review? Did anyone in security know how many humans and machine credentials could write to that field? I would bet real money that at most of the 178, the honest answer is no. It was never classified as privilege, so it was never governed as privilege.

Look at what Report URI told site owners to do after the fact. Open the Project JavaScript in every project. Review which users, API tokens and integrations can write to your account, and rotate their credentials [3]. That is an access review. It is a joiner, mover, leaver and secrets rotation exercise for a marketing tool, run in a panic because nobody ran it calmly before.

And the vendor side of the same question is ugly. Report URI said it did not know how the attacker obtained write access to so many unrelated customer projects, adding that was for Optimizely to explain [3]. When Smartsheet’s script was cleaned once and then re-infected with a hidden version, the researchers read that as a sign the attacker still had access at that point [3]. Write access across tenants, persisting through cleanup. In any other context we would call that what it is: privilege escalation inside the vendor’s identity plane.

This angle is related to an earlier piece of mine about a SaaS app vendor holding tokens for every merchant at once, but it is not the same argument. That one was about aggregation, about one vendor sitting on everybody’s keys. This one is about classification. The privilege in the Optimizely case was never stolen from a vault. It was granted in the open, through a sign-up form and a seat license, by people who would never describe what they were approving as code execution.

It was not just one vendor

Help Net reports that Hudson Rock and an independent researcher traced poisoned scripts to three more services, Kissmetrics, StorifyMe and Minute Media, with earliest tampered responses recorded on November 13 [4]. The researchers are careful to say shared hosting and loader code does not prove a single operator or a single way in [4], and I am going to be equally careful. But the pattern is obvious even without attribution. The median homepage loads at least 22 JavaScript files, per the HTTP Archive figure Krebs cites [1]. Every one of those is a vendor with some set of humans and tokens that can change what runs on your page.

The Register closed its story with a line that stung because it is true: all that money spent securing corporate websites, and the attackers came in through software the marketing department uses to tinker with pages [2].

The counterargument, and why it only goes so far

Let me give the fair version of the other side. Cynet’s researchers point out that a site that delivered the code has not necessarily had its own systems compromised [3]. That is accurate, and it matters for breach disclosure and for not panicking customers. Card details entered into a payment provider’s cross-origin iframe were out of the skimmer’s reach, and cookies marked HttpOnly were safe from theft [3]. Good engineering hygiene limited the blast radius for a lot of people.

But “our infrastructure was not breached” is cold comfort to the visitor whose clipboard got rewritten on your checkout page. Your brand served the payload. Your origin executed it. The browser does not care whose servers were clean.

What I would actually do on Tomorrow

Start by inventorying every third-party tag that can change executable content on your pages, and for each one, count the identities that can write to it. Humans, agency accounts, service accounts, API tokens, integrations. If you cannot produce that number, that is your finding.

Then treat write access to those fields as privileged access. Named owners. Phishing-resistant authentication on the vendor console. No shared agency logins. Tokens scoped to the narrowest project and rotated on a schedule, not after an incident. Remove access when the campaign ends or the contractor leaves, because standing write access to customer browsers is standing privilege by any honest definition.

Keep these tags off checkout and authentication pages entirely, which is exactly what StepSecurity told affected teams [3]. Then write your CSP around behavior, not hostnames: tight connect-src, no eval, no blanket trust of a vendor CDN, because that CDN is only as trustworthy as the least-protected account that can publish to it.

Finally, ask your vendors the question Report URI asked Optimizely in public: who, inside your platform, can change what runs on my site, and how would I know [3]?

The lesson underneath

Trust is not a property of a domain name. It is a property of the identities that can change what that domain serves, right now, today. Helme put it plainly: trusted JavaScript does not always remain trustworthy [1]. I would go one step further. It was never the JavaScript you trusted. It was every person and every token behind the pen, and you never met most of them.

About the Author

Len Noe is a Solutions Architect at BeyondTrust, Transhuman, Podcaster, International Cyber Security Speaker, Author, Technical Evangelist, and Biohacker with 11 implanted microchips. A former blackhat and ex-1% motorcycle club member with more than 30 years in technology, he has presented in over 70 countries and is featured in the documentary I Am Machine, which premiered at DEF CON 2025.

References

  1. Brian Krebs. “Attackers Used Optimizely to A/B Test Malware on Top Websites.” Krebs on Security, November 17, 2026. krebsonsecurity.com
  2. Carly Page. “Optimizely hack turns world’s biggest websites into malware dispensers.” The Register, November 17, 2026. theregister.com
  3. Swati Khandelwal. “Hijacked Optimizely Scripts Served Card Skimmer and iOS Exploit Kits to CNN, BBC and Zoom Visitors.” The Hacker News, November 17, 2026. thehackernews.com
  4. Sinisa Markovic. “Malware spread through Optimizely hits 178 organizations, including BBC, CNN and Zoom.” Help Net Security, November 17, 2026. helpnetsecurity.com