⏱️ Reading time: 7 min
On November 16, 2026, someone wrote malicious code into the custom JavaScript field of hundreds of Optimizely projects, turning an A/B testing tool into a malware distribution channel. Security firm Report URI documented the attack and counted 178 affected organizations, including Zoom, the BBC, CNN, Salesforce, GitLab, IKEA, HSBC, Equifax, Air France-KLM, Ledger and Peacock.
📑 En este artículo
Here’s how Optimizely Web works: each customer loads a JavaScript snippet from the company’s CDN, for example cdn.optimizely.com/js/6395285578973184.js, and that file decides which page variant to show each visitor. The snippet includes an optional field called Project JavaScript: custom code that the customer writes from their Optimizely dashboard and that runs on every page load. The attacker wrote their own code there, left the rest of the snippet untouched, and never touched the certificate or the CDN infrastructure. That’s why every affected site kept serving the malware from a domain and certificate it already trusted, according to Report URI’s report.
What the script did on every visit
The payload, a file called track.js, performed several functions at once and was rewritten 26 times over the course of the day: the attacker kept adding new functions and wallets but never removed any, according to Report URI’s tracking. Its most direct function scanned page text, forms and the clipboard for cryptocurrency addresses (BTC, ETH and EVM chains, TRON, Solana, TON, Litecoin, Bitcoin Cash and Dogecoin), validated the checksum to confirm it was a real address, and replaced it with one of the attacker’s own.
The same script stole payment card data, cookies and screenshots. On iPhone and iPad visitors running iOS 18.6 or earlier, it injected an iframe that delivered the exploit kits known as Coruna and DarkSword. Report URI confirmed cryptocurrency transactions that had already completed into the attacker’s wallets.
The attacker used three loader variants to avoid relying on a single method: a direct script tag to 1688360.cc/track.js in most snippets, the same URL hidden in base64 inside an atob() call so a simple domain search would not find it, and a full inline copy of the payload that runs via eval if the 1688360.cc domain is blocked. That last variant makes blocking the domain alone useless: the code is already inside the page.
flowchart TD
A["Attacker"] --> B["Project JavaScript field in the Optimizely dashboard"]
B --> C["Snippet served from cdn.optimizely.com"]
C --> D["Visitor's browser"]
D --> E["track.js: wallet swapping and card theft"]
D --> F["iOS 18.6 or earlier: iframe with Coruna and DarkSword"]
Optimizely attack timeline: from a clean snippet to 186 infected
The first confirmed injection into an Optimizely snippet happened at 04:21 UTC on November 16, in the code loading the BBC; at 02:21 that same copy was still clean. Minutes later, certificates were issued for the 1688360.cc domain, registered at 05:23, and the first version of track.js appeared at 08:10. Cryptocurrency theft toward the attacker’s exchanges was logged at 13:03, and card theft was added at 13:43.
Mass cleanup did not start until between 20:40 and 21:10 UTC: Ledger at 20:41, Smartsheet at 20:56 and the BBC at 21:07. By 21:52, only 1 of 188 snippets remained infected (TalkTalk’s), down from 169 an hour earlier, according to Report URI’s tracking. One detail a quick read of the case tends to miss: Smartsheet’s snippet was cleaned once and then reinfected with the base64 loader version, which indicates whoever was writing the code still had access during the incident, not only at the initial moment.
What this means for you if your site uses experimentation tools
If your product or marketing team uses Optimizely, or any platform that lets you write custom JavaScript inside a third-party snippet, check today who has write access to that project. The attack did not break Optimizely’s CDN: it got in through access to specific customers’ configuration. If your snippet is served from your own domain as a proxy, as some victims did with addresses like opt.scheels.com, you are just as exposed, because the content comes from the same Optimizely project.
The second action is technical: turn on a Content Security Policy in enforce mode (not just report-only) that limits which domains your experimentation provider’s script can connect to. A correct CSP would not have blocked the snippet itself, since it still came from cdn.optimizely.com, but it would have cut the outbound connection to 1688360.cc if that domain was not on the allowlist. Report URI makes exactly this point: its monitoring of unexpected changes in trusted JavaScript would have shown the snippet’s alteration even though it came from a trusted domain.
⚠️ Heads up: if your Optimizely project serves the snippet from your own domain as a proxy, the attack affects you just as much as anyone loading it directly from cdn.optimizely.com: the content is the same.
Optimizely vs. LaunchDarkly, Split and Statsig
Optimizely Web is not the only way to decide which version of a page each user sees. LaunchDarkly and Statsig resolve feature flags and experiments by evaluating rules inside an SDK that runs in the customer’s own code, without an open field for pasting arbitrary JavaScript that the provider serves on your behalf. Split, now part of Harness, follows a similar model: the customer’s own code decides which variant to show, and the platform only delivers the configuration.
The difference is not that these alternatives are immune to compromise (any third-party SDK is an attack surface), but that none of them offers, by design, a Project JavaScript style field where a single dashboard account can write code that runs unreviewed on every page load across hundreds of different customers. That field is exactly what the attacker used in Optimizely.
What the community is saying
In the Hacker News discussion about the case, user kursus pointed out that before the concept of a tag manager existed, it was common for sites to blindly load a JS file from their tracking provider’s server, maintained by that provider’s team rather than the site’s own. For kursus, trusting a third-party script without reviewing it is not a new practice, it just changed names. Another comment, from dboreham, speculates that some large sites like CNN might separate their paywalled pages from marketing and tracking scripts, though they note there is no confirmation that was the case in this attack.
The limits of Optimizely’s response
Optimizely has not publicly explained how the attacker got write access to so many unrelated customer projects. It could have been a compromised internal system, a credential with cross-account access, or something else the company has not yet detailed. Report URI clarified that this was not a CDN cache poisoning attack: the CDN served exactly the content each project had configured.
Cleanup was fast once detected, from 169 infected snippets to 1 in under an hour, but it does not close the underlying problem. If the access came from a stolen credential or an internal Optimizely system, nothing prevents reinfection until the company confirms and fixes that path. There’s also another CSP detail that the eval-based payload mode makes relevant: a policy that still allows unsafe-eval would not have blocked that inline copy, only the two versions that depended on loading an external script from 1688360.cc. Removing unsafe-eval from a CSP is harder than adding a domain to a blocklist, but it is the only real defense against that third mode.
References
- Report URI: World’s Largest Brands Infected With Malware Via Optimizely: full timeline of the attack, list of 178 affected organizations and the injection mechanism in the Project JavaScript field.
- Report URI: technical detail of the track.js payload, including cryptocurrency theft, card skimming and delivery of the Coruna and DarkSword exploit kits on iOS.
- Report URI: recommendation to monitor changes in trusted JavaScript and use a CSP in enforce mode to mitigate similar attacks.
- Hacker News: community discussion of the incident and its implications for sites loading third-party scripts.
Featured image: Foto de Kevin Horvat en Unsplash
How each article is made (sources, testing and editorial criteria): How we work.
Did it work for you? Got a different error? Say so below: questions get answered and help the next reader.
Leave a comment
0 Comments