On 16 November, visitors to the websites of 178 organisations were served malicious JavaScript. The list includes Zoom, the BBC, CNN, Salesforce, GitLab, IKEA and HSBC. None of them were breached in the usual sense. The code came from cdn.optimizely.com, with a valid certificate, from a vendor every one of those sites had chosen to trust.
What’s striking is how simple the route was. Optimizely Web is an A/B testing platform. Each customer loads a snippet from Optimizely’s CDN, and that snippet includes a “Project JavaScript” field, custom code that runs on every page. The attacker wrote their code into that field and left the rest alone. Optimizely says an unauthorised party exploited a vulnerability in its Experimentation product to do it, and that the flaw is fixed and the injected code removed.
Researcher Scott Helme of Report URI found 188 infected snippets. By his analysis the code swapped cryptocurrency wallet addresses, later added a card skimmer, and tried to load exploit kits on older iPhones. These are early details, and I’ll update this post if they change.
Update, 18 November: nobody has published an estimate of how many people were exposed. The 178 counts organisations, not visitors, and Report URI says the real number is likely higher. The only confirmed loss is about $7,150 in crypto sent to the attacker’s wallets. That figure leaves out skimmed cards, stolen cookies and infected iPhones, because there’s no visibility into those yet.
Same family of problem
It’s a supply chain attack, and a sharper one than the node.js cases. It’s the same family as the inherited-code risk I wrote about in 2022. In September 2025, chalk and debug were poisoned with a browser-side crypto clipper, the same trick used here. In 2024, the polyfill.io domain changed hands and started serving malware to more than 100,000 sites. And Bybit was JavaScript tampered with at a vendor, then served to a browser.
The difference is when the trust is exercised. With npm, the poison has to get into a build, where you can pin versions, use a lockfile and scan what you ship. Here nothing changed on the customer side: no deploy, no pull request. Someone else’s server decides what your visitors run, on every page load. A dependency you can’t version is a dependency you don’t control.
What everyone can do in their browser
A visitor has limited control here. These were sites you’d have had no reason to avoid. But three things shrink the blast radius:
- Keep your browser and operating system patched. Report URI’s advice for the iPhone exploit kits was iOS 18.7.3 or later, or iOS 26.3 or later.
- Run a content blocker, and keep other extensions few. A blocker won’t stop code from a host the site trusts, but blocklists catch the attacker’s own domains once they’re known. Extensions are inherited code too.
- Protect the money. Check crypto addresses after you paste them, on a hardware wallet’s screen if you have one. Pay with a wallet or virtual card, since cards in hosted payment iframes were out of the skimmer’s reach, and check your statement if you used a card on an affected site on 16 November.
What I’d ask on Monday
- What third-party code runs on our login and payment pages, and who can change it? Build the inventory, including what those scripts load in turn. A tag manager or experimentation console is a deployment pipeline into production, so treat its write access that way. This time the flaw was on the vendor’s side, so ask how they protect theirs.
- Does our CSP contain, or just allow? Allowing cdn.optimizely.com is the obvious thing to do, and it didn’t help, because the code came from that host. Only 24 sites, about 15% of those Helme assessed, had a policy strict enough to block it. On 17 more the payload ran, but a tight
connect-srcstopped the data leaving, although the address swapping still worked. - Does a marketing tag need to run where card details are typed? Keep third-party tags off login and payment pages, use hosted payment fields, and set
HttpOnlyon session cookies. If you take cards, PCI DSS 6.4.3 and 11.6.1 already ask you to inventory payment-page scripts and detect changes. - Would we notice, and could we switch it off? Monitor third-party scripts for changes. Helme’s company sells exactly that, so weigh the source. Then test the other half: can we pull a tag in minutes, and who tells customers? The first injection was seen at about 04:21 UTC, and most of the cleanup came that evening.
Nobody on those 178 sites wrote a line of the code that swapped wallet addresses. They inherited it, along with the trust. Worth knowing what else is in the estate.
More information
- Report URI – World’s Largest Brands Infected With Malware Via Optimizely (Scott Helme’s analysis; Report URI sells script monitoring)
- Help Net Security – Malware spread through Optimizely hits 178 organizations
- Sophos – Hijacked Optimizely snippets served malicious code
- StepSecurity – OPTIMJACK write-up
- The Hacker News – npm chalk and debug compromise
- BleepingComputer – Polyfill.io supply chain attack impacts over 100K sites
- Report URI – PCI DSS 6.4.3 and 11.6.1 client-side security guide (vendor guide)
