
Optimizely Supply Chain Attack: How to Check If Your Sites or Users Were Exposed
The Optimizely supply chain attack was a November 16, 2026 compromise in which an attacker exploited a flaw in Optimizely Web Experimentation and added malicious JavaScript to some customers' projects. For most of that day, visitors to websites run by at least 178 organizations, including the BBC, Zoom, Salesforce and GitLab, received the code inside the Optimizely snippet those sites already trusted.
Depending on the page and device, the code swapped crypto wallet addresses, skimmed card details from a site's own payment forms or tried to load exploit kits onto older iPhones. Report URI, which found it in 188 of the 1,611 Optimizely snippets it checked, published the discovery the same day.
The question leadership will ask is whether the organization was affected. Answering it takes a website check, run by the team that manages your Optimizely account, and an employee check, run by the SOC.
A company that has never used Optimizely still needs the employee check, since employees may visit affected sites such as zoom.us and gitlab.com for work. That check is a log search, the kind of hunt an AI threat hunting agent can run across every connected source without anyone writing queries.
Prophet AI Threat Hunter has an emerging-threat hunt for the Optimizely supply chain attack, ready to run or schedule.
The checks below cover the Optimizely wave, and the employee check also looks for the related injections that began on November 13. Poisoned npm and PyPI packages leave different evidence and are covered in our guide to supply chain attack detection.
{{ebook-cta}}
What happened in the Optimizely supply chain attack
Optimizely Web Experimentation lets a company test changes to its website without changing the site's code. Each site loads a JavaScript snippet, usually from Optimizely's CDN, and the snippet includes an optional field called Project JavaScript, where the site's team can add code of its own. The attacker put a loader in that field, so an infected snippet still came from its usual address with the rest of its contents untouched.
The timeline below uses UTC and comes from Report URI and StepSecurity. All times are on November 16 unless noted.
- 04:21: the first injection is seen, in the BBC's snippet, loading a script from
webshell[.]lol. - 05:23: the payload domain,
1688360[.]cc, is registered. - 13:43: a card skimmer is added to the payload.
- About 20:40 to 21:10: most infected snippets are cleaned.
- 21:52: one of the 188 infected snippets still carries the code.
- 22:47: Report URI publishes its findings.
- November 17, 01:02: Optimizely posts an incident notice.
The payload, a script called track.js, was rebuilt 26 times during the day, according to Report URI. In the versions researchers analyzed, it did four things:
- Swapped valid cryptocurrency wallet addresses for the attacker's own, on the page, in the clipboard and in the page's outgoing requests.
- Captured card numbers, expiry dates and security codes typed into the site's own forms, though not payment fields hosted in another provider's iframe.
- After a wallet swap near payment-related words, sent a screenshot of the page and any cookies without the
HttpOnlyflag to the attacker. - On iPhones and iPads, loaded a second-stage script from a set of
.icuhosts that route older iOS versions to the Coruna and DarkSword exploit kits.
Optimizely says an unauthorized party "exploited a vulnerability in Experimentation" and that it has fixed the flaw, removed the injected code from affected projects and taken steps to secure affected user accounts. Press reports, including The Register's, place the flaw in the account sign-up process. As of November 17, Optimizely had not said how a sign-up flaw gave the attacker write access to other customers' projects.
As of November 17, no published research showed an iPhone compromised by this attack. Hudson Rock says its iOS findings show the exploit kits being delivered without demonstrating a compromised device. No researcher or vendor had named the actor behind the campaign as of November 17, though some researchers call the campaign OPTIMJACK.
Is the Optimizely hack linked to the Kissmetrics, StorifyMe and Minute-ly.com injections?
The two waves share infrastructure, and Report URI calls webshell[.]lol "the same actor's injector," though no one has publicly shown that one operator ran both. From November 13, the same first-stage loader, webshell[.]lol/get.js, turned up in scripts served by Kissmetrics, StorifyMe and Minute-ly.com, according to Hudson Rock. Hudson Rock found the BBC's injected Optimizely snippet loading the same file early on November 16, with both waves sending iPhones to the same exploit hosts.
Hudson Rock also notes the limits of that evidence. Shared loaders and hosts show linked delivery activity, but they do not show that one person controls every affected vendor. Its research also does not establish how the attacker got into each vendor.
Why the payload got past most Content Security Policies
A poisoned package, like the malicious Axios releases on npm, reaches a company through its own installs and builds. Optimizely edits usually reach visitors without passing through the site's deploy pipeline or code review, which is how the product lets teams change a site without touching its code. The attacker's edits went live the same way.
A Content Security Policy that trusts the snippet, whether by host or by nonce, cannot tell the attacker's Project JavaScript from the rest of it. Both arrive in the same file. What a policy could still control was the next step, when the loader fetched track.js from 1688360[.]cc and the payload sent stolen data back to the same host.
![Diagram of the Optimizely payload's path through a visitor's browser and where a Content Security Policy could act. The attacker's loader sits inside the Optimizely snippet, so a policy that trusts the snippet can't separate the loader from it. The loader either fetches track.js from 1688360[.]cc, which a strict script-src host list blocks and 'strict-dynamic' lets load, or runs an inline copy with eval, which a script-src without 'unsafe-eval' or 'unsafe-inline' blocks. Once the payload runs, its wallet swap in the page and clipboard still runs where connect-src blocks the theft; card details sent to /card.php and a screenshot and cookies sent to /hit.php after a swap are blocked by a strict connect-src; and on older iPhones and iPads, exploit kits load from .icu hosts in a hidden iframe, which a strict frame-src can block but connect-src doesn't cover.](https://cdn.prod.website-files.com/66096b1ab0969e1098f75192/6afcec755e6c2fde1884aef1_AIpPl2DaE6gr1dpdaVI1EFQGtex7h56vE_WKQJkF-4ypCWSz50GDp6y8k_dpWcSRSNNM5etRt29pw81f83I10vsrd7_n8VhIagQU9wnu5PlZJYFOH92DtfuDeX6cQ0ClgWhWS4Ht1AgGi5i-fF-8h26BHBvkdEknmlExngx4VjDMvfharnGGReo%3Ds2048.png)
A strict script-src that named only a site's known hosts blocked that fetch. A strict connect-src blocked the data theft, though the wallet swap still ran in the page. connect-src also does not govern frames, including the hidden iframe that loaded the iPhone exploit kits. Two settings let the payload through anyway, according to Report URI.
'strict-dynamic' lets a script the policy already trusts add other scripts to the page, wherever they are hosted, so the trusted snippet could pull in track.js. Optimizely's own CSP guidance pairs a nonce with 'strict-dynamic', a setup often recommended for blocking cross-site scripting. 'unsafe-eval', which Report URI says Optimizely's documentation told customers to allow, let through a version of the loader that carried the whole payload inline and ran it with eval.
Report URI checked the policies on affected organizations' own sites. It found no policy at all on 68, and a policy that let both the payload and the data theft through on 50. On 17 more, connect-src stopped the theft but not the payload, and only 24 sites, about 15 percent, blocked the payload outright.
Did your Optimizely projects serve the injected code?
Run this check if your company uses Optimizely Web Experimentation anywhere, including on sites an agency or subsidiary manages. Because Optimizely has removed the injected code, a snippet you fetch today will likely look clean. Evidence of what visitors received comes from Optimizely's records, your own logs, public scans of your site from November 16 and the infected-snippet archive Report URI offers affected organizations.
Start with each project's change history from November 15 through today. According to Optimizely's documentation, it records every change to Project JavaScript and to variation and shared code. It also shows who made each change and whether it came through the web interface or the REST API.
A change there that nobody on your team or at your agencies made is strong evidence your project served the code. Confirm each change with the person the history names, since Optimizely says some user accounts were affected, and an edit made through one of them would carry that user's name. A clean history is weaker evidence, because Optimizely has not said whether changes made through the flaw were recorded. Ask Optimizely which of your projects were modified, when, and through which accounts. Whatever the answer, review the users, API tokens and integrations that can write to your account, as Report URI advises every site that uses Optimizely Web Experimentation.
Then search Project JavaScript, every experiment's custom code and any shared code for 1688360, webshell, atob( and eval(, and read any code nobody on your team wrote. Other code also ships in the snippet, such as experience templates used in variations, conditional activation code and enabled custom analytics integrations, so search it too. Report URI found three versions of the loader:
- A plain script tag for
1688360[.]cc/track.js - The same address hidden in base64 and decoded with
atob(), which a search for the domain misses - A full inline copy of the payload, run with
eval, which blocking the domain does not remove
Repeat that search for the next several days. According to The Hacker News, researchers saw at least one snippet infected again with the base64 version of the loader after it had been cleaned. That suggests the attacker still had access at that point.
If you serve the snippet from your own domain or CDN, check that copy too. Report URI found self-hosted snippets carrying the code, since they take their content from the same Optimizely project. Depending on how your copy syncs and caches, it may have kept serving the injected version after Optimizely cleaned its own CDN.
If your Content Security Policy sends violation reports, look for blocked requests to webshell[.]lol, 1688360[.]cc or the .icu hosts. A report like that shows a page ran the attacker's loader and the policy blocked that request. Where the policy also allowed 'unsafe-eval', the inline copy of the payload could still have run.
A report-only policy sends the same reports without blocking anything. A policy that allowed everything the payload did sent no report, so a quiet report log does not clear you. On payment pages, also check the alerts from the change- and tamper-detection mechanism that PCI DSS requirement 11.6.1 calls for.
Last, work out what the snippet could reach. A snippet on checkout or payment pages put card details typed into your own forms within the skimmer's reach. A cookie's Path attribute is not a security measure, so code on any page of a domain can usually read that domain's cookies that lack the HttpOnly flag.
Did your employees load the injected code from other sites?
Run this threat hunt across proxy, DNS, secure web gateway and EDR network logs, starting at 04:00 UTC on November 16, as StepSecurity recommends. Some summaries give 08:00, close to the first recorded request to 1688360[.]cc, but the first injection, at 04:21, loaded webshell[.]lol instead.
For webshell[.]lol and the .icu hosts, search all the retention you have, since both were in use before the Optimizely wave. Run every search through to today, and add the domains below to your DNS and proxy blocklists, since some of the .icu hosts still resolved on November 17. These indicators come from StepSecurity's write-up, defanged here; its two hashes match two of the payload's 26 builds, so search on the domains first:
1688360[.]cc: the payload and data theft host, with the paths/track.js,/card.php,/hit.phpand/html2canvas.min.js136.113.106[.]136: the last address1688360[.]ccresolved to, a Google Cloud address that can be reassigned once released, so a later hit on it alone may be unrelated. The same holds for34.87.36[.]58and34.96.213[.]201, the Google Cloud addresses that hosted five of the seven.icudomains StepSecurity lists belowwebshell[.]lol: the first-stage loader host, path/get.jsf7b84b9a94f749f8ced3f0112d0a66894e8bbc3b4aa5592021d1b8976a9bad40: SHA-256 of the finaltrack.js617df364120adc47fb4f55dcc0c69a9b6c9e108a92c60bc2b64a266254ab02f3: SHA-256 of the first capturedtrack.js
Hudson Rock records ios-hack[.]icu and je4ft3gn8vxudum[.]icu from the iPhone stage's requests and code, and StepSecurity lists these .icu hosts:
5vobauua3ojpklx[.]icup55abnw2pnsz0bo[.]icueil786npto9u5bq[.]icu1y69gy2mdzaltz1[.]icugm38duxoihbmy5o[.]icul7bcsnwc7nzr3uo[.]icuq1ls1ho9x7wsxlu[.]icu
Indicators like these age quickly. 1688360[.]cc was placed on registry hold early on November 17, and a campaign that continues can move to new hosts. A hit or a clean result answers the question only for the indicators listed here.
To catch domains the campaign registers next, search the same logs for requests to newly registered domains, which many web gateways and DNS filters can flag. The payload domain, 1688360[.]cc, was in use about three hours after it was registered. Newly registered domains also include legitimate new sites, so start with the requests a device made right after it loaded one of the affected sites.
Set aside hits from your own analysts and security tools looking these indicators up, then for each remaining hit, establish the user, the device and the referring site. Where your proxy inspects TLS and logs the referrer, it shows which affected site served the code. Where it does not, compare the sites the device looked up or requested just before the hit with the organizations Report URI lists as affected. From the device type, you can tell which parts of the payload could have reached it. If your proxy records full URLs, an allowed request to /card.php means card details were sent. An allowed request to /hit.php means a screenshot and the site's cookies without the HttpOnly flag were sent after a wallet swap, so treat the user's sessions on that site as exposed.
On laptops and desktops, none of the published analyses describes the payload installing anything. It worked inside the browser tab. The exposure comes mostly from what the user did while that page was open, such as typing card details into the site's own payment form or copying a wallet address anywhere. On a page that showed a wallet address near payment wording, the payload could also swap it and send the /hit.php request with no action from the user.
For iPhones and iPads that reached the .icu hosts, check in your MDM which iOS version each one ran on the day it reached them. The user agent in your logs is a weaker guide, since Safari on iOS 26 reports a frozen, older iOS version. Hudson Rock found the loader routing iOS 13 to 17 to the Coruna exploit kit and iOS 18 to a chain labelled DarkSword. Report URI advises updating to iOS 18.7.3 or later, or iOS 26.3 or later, so investigate devices below 18.7.3, and iOS 26 devices below 26.3, first.
Your logs cover the devices your proxy, DNS service or EDR can see. A personal phone browsing over cellular data leaves nothing in them, so a clean result says nothing about devices you don't monitor. Safari traffic and DNS lookups sent through iCloud Private Relay can also bypass those logs on your own network, unless the network blocks the relay.
What to do if a check finds a hit
If your project served the code, preserve the evidence first: save the change history, CSP reports and CDN logs. If a snippet you serve still carries the code, save a copy and take the snippet off your pages until it is clean. Then, from an Optimizely Administrator account, remove users, API tokens and integrations you don't recognize, and rotate the credentials of the rest. After that, confirm the code is gone from every field in the project that holds JavaScript, from the snippet Optimizely's CDN serves and from any self-hosted copy.
If your own forms take card details, Report URI advises treating cards entered on November 16 as potentially exposed. Bring in your payment processor, fraud team and legal counsel. Decide with legal whether to tell customers your site served them malicious code.
Report URI also advises treating session tokens in cookies without the HttpOnly flag as exposed and rotating them. End the old sessions on the server when you do, since a stolen session token usually keeps working until it expires or the server revokes it. Watch the affected accounts for signs of session hijacking afterward.
For an employee with a hit, cancel and reissue any card entered on an affected site between about 04:00 and 21:00 UTC on November 16, as StepSecurity recommends. Extend that window to the time of the hit if it came later, since some snippets still carried the code after 21:00. Check any crypto transfer to an address shown on, typed into or copied from one of those sites, or copied anywhere else while one was open in a tab. Update iPhones and iPads below the fixed versions, and keep them off corporate resources until they have been reviewed. In the DarkSword attacks Lookout analyzed, the kit took saved passwords and cookies within minutes and then deleted its staged files, so a later update does not undo a theft. For a device below the fixed versions that reached the .icu hosts, also reset the passwords and revoke the sessions of the corporate accounts used on it.
If every check comes back clean, the hardening still applies. Report URI recommends script-src and connect-src lists that name only the hosts a site needs, avoiding 'unsafe-eval' where possible, and turning on violation reporting. StepSecurity also advises keeping third-party JavaScript off checkout pages where you can, and taking card details through payment fields hosted in a provider's iframe.
Moving off 'strict-dynamic' usually means listing every host your third-party scripts load from, which takes testing on a site with many tags. A host-list policy also protects less against cross-site scripting, since such policies can often be bypassed. A site that keeps 'strict-dynamic' for that reason can still block this payload's data theft with a strict connect-src, which does not govern script or image loads. Optimizely's notice lists the steps it has taken itself but, as of November 17, none for customers.
How Prophet AI checks for exposure in your logs
The website check runs on Optimizely's records, so it stays with the team that manages your Optimizely account. The employee check is a hunt across your logs, and that is the part Prophet AI runs.
Prophet AI Threat Hunter turns a plain-language question and an indicator list into a hunt across the sources a team has connected, such as proxy, DNS, EDR and identity logs. The team does not need to learn a query language. For severe emerging threats like the Optimizely supply chain attack, those hunts run automatically in the background, and the team sets the scope and cadence of its own hunts.
Suspicious leads go to Prophet AI SOC Analyst for a full investigation that records every query and piece of evidence. An analyst can check that work instead of redoing it. What the investigation confirms becomes a detection opportunity for Prophet AI Detection Engineer, backtested against the team's own history before anyone approves it.
If the employee check is still open, see how Prophet AI Threat Hunter runs hunts for emerging threats.
The Optimizely supply chain attack was a November 16, 2026 compromise in which an attacker exploited a flaw in Optimizely Web Experimentation to add malicious JavaScript to some customer projects. The code then reached visitors of websites run by at least 178 organizations through the Optimizely snippets those sites load, and it swapped crypto wallet addresses, skimmed card details from the sites' own payment forms and tried to load exploit kits onto older iPhones. Some researchers call the campaign OPTIMJACK.
Check each project's change history for edits to Project JavaScript or custom code from November 15 onward that your team did not make, and ask Optimizely which of your projects were modified and through which accounts. Then search your project code, including any self-hosted copy of the snippet, for 1688360, webshell, atob(, eval( and code nobody on your team wrote, and review CSP violation reports for blocked requests to the attacker's hosts.
Search proxy, DNS, secure web gateway and EDR network logs for 1688360[.]cc from 04:00 UTC on November 16 onward, and for webshell[.]lol and the attack's .icu hosts across all the retention you have. For each hit, identify the user, the device and the referring site. A company that does not use Optimizely still needs this check, because the affected sites included Zoom, Salesforce and GitLab.
No compromised iPhone had been publicly confirmed as of November 17. The payload tried to load the Coruna and DarkSword exploit kits on older iPhones and iPads, and Hudson Rock says its findings show the kits being delivered without demonstrating a compromised device. Devices below iOS 18.7.3, or on iOS 26 releases below 26.3, that reached the attack's .icu hosts are the ones to review first, and Report URI advises updating them.
One that limited scripts to named hosts could stop the payload. Report URI found that 24 of the affected sites it checked blocked the payload with a restrictive script-src, and 17 more blocked the data theft with connect-src. Policies using 'strict-dynamic' let the payload run, and 'unsafe-eval' let an inline copy of it run. A policy that allows Optimizely's CDN could not block the injected loader itself, because it arrived inside the trusted snippet.
A package compromise reaches an organization through software it installs or builds, so the evidence sits on laptops, CI runners and build logs. The Optimizely attack reached website visitors through a vendor's snippet, usually from its CDN, with no install or build involved, so the evidence sits in Optimizely's change history, Content Security Policy reports and proxy logs. Prophet Security's supply chain attack detection guide covers package compromises.
Insights
Definitive Guide to Detection Engineering
Where detection engineering falls short, how to close those gaps, and what it looks like in an AI-enabled SOC.



.avif)
