SecuritySupply ChainJavaScriptContent Security PolicySaaS

Optimizely Supply-Chain Attack: How to Check Your Site and Harden Scripts

person
LaunchBoosts Research Desk·AI-assisted research
9 min read

Researched and drafted with AI assistance from the 6 public sources listed at the end of this article, then published after automated editorial checks. Spotted an error? Tell us at support@launchboosts.com.

Optimizely Supply-Chain Attack: How to Check Your Site and Harden Scripts — LaunchBoosts

On November 16, 2026, an attacker got into Optimizely Web, the A/B testing platform, and used it to push card-skimming, crypto-stealing and iPhone-exploit JavaScript to visitors of 178 organisations, including Zoom, GitLab, the BBC, CNN and Salesforce. Optimizely says the injected code has been removed. Every team that loads a third-party tag still has two jobs: confirm their own snippet is clean and keep evidence, then change their setup so the next compromised vendor can't run code on checkout and login pages. This guide covers both.

What happened, in order

Optimizely confirmed to BleepingComputer that "an unauthorized party exploited a vulnerability in the Experimentation account signup flow" and "used it to add malicious JavaScript to some customer projects." The company says it closed the vulnerable path, removed the injected code, terminated active sessions, revoked tokens, reset affected accounts and blocked delivery of the code through Cloudflare. As of 01:02 UTC on November 17, the Optimizely status page listed the incident as "Monitoring". It says the vulnerability is fixed and "additional protections" are in place "to prevent unauthorized code from being delivered."

The detailed timeline comes from Scott Helme's team at Report URI. All times are UTC on November 16 unless noted:

Time Event
Sep 17 The same actor's injector (webshell.lol) first seen on Turkish sites
Nov 13–16 Earlier wave injects webshell.lol/get.js through Kissmetrics, StorifyMe and Minute Media scripts
04:21 First Optimizely injection observed (BBC snippet)
04:28–05:23 Payload domain 1688360.cc registered and certificates issued
08:10 First build of track.js, a crypto clipper
11:20 Zoom's snippet modified
13:43 Card skimmer added
17:13–17:57 iOS exploit kit components added
19:24–19:59 Report URI's sweep finds infections in 186 of 1,611 snippets
~20:40–21:10 Mass clean-up begins
21:52 One snippet still infected

The attacker didn't touch Optimizely's core library. They wrote code into each customer's Project JavaScript field, a custom script that runs on every page where the snippet loads. According to Report URI, the CDN "served exactly what it was given." This wasn't cache poisoning. The files were legitimate, signed-in-the-trusted-place files containing malicious code. Report URI says the attacker rebuilt the payload 26 times that day.

How many sites, really?

Counts differ because each researcher looked at a different sample:

  • Report URI: 188 infected snippets across 178 organisations, out of 1,611 it could fetch. It says the true number is likely higher.
  • BleepingComputer: independently confirmed through urlscan.io that the script loaded or tried to load on 21 named sites. These include Zoom, KLM, GitLab, BBC, CNN, Salesforce, Fox News, HSBC Insurance, Bitly, Smartsheet and UNICEF USA. Salesforce disabled Optimizely scripts across preview and production as a precaution.
  • Amirul Cyber, which named the campaign OPTIMJACK, described about 15 affected sites. It also found that some of them, including Fox News, American Eagle and Air France, were hit through different delivery paths: an ad wrapper, a first-party bundle and an obfuscated script.

Treat all of these figures as counts of affected sites, not victims. Infostealers.com notes that "the number of malicious page loads, successful thefts and compromised devices remains unknown." Report URI and BleepingComputer both put the money received by the attacker's wallets at about $7,000. Some of those wallets were active before the attack, so the amount actually stolen is unclear.

What the payload did

The researchers broadly agree on four modules:

  1. Crypto clipper. It rewrote wallet addresses in page text, form fields, the clipboard and outgoing requests, and validated real checksums so the swapped addresses looked legitimate.
  2. Card skimmer. It captured card number, expiry and CVV from on-page forms and sent them to 1688360[.]cc/card.php. Cards entered inside hosted payment iframes were out of reach, according to Report URI.
  3. Cookie, URL and screenshot theft. It took non-HttpOnly cookies and page captures, using html2canvas loaded from the attacker's host.
  4. iOS stage. On iPhone and iPad it loaded the Coruna and DarkSword exploit chains in a hidden iframe. Infostealers.com stresses that these were observed deliveries in a headless scanner, not confirmed device compromises.

Two details matter for defenders. First, some injections used Base64 and atob() rather than a plain URL. Second, some variants embedded a full copy of the payload inline for when the remote script was blocked. Blocking the domain alone doesn't remove the threat.

How to check your own site today

If you use Optimizely Web, or any tag that lets people edit JavaScript in a dashboard, work through this list:

  1. Check for Optimizely's notification. It told BleepingComputer it notified all affected customers. A missing email isn't proof you're clean, so do the checks below as well.
  2. Save evidence before you change anything. Infostealers.com recommends keeping response bytes, hashes, headers and timestamps of the served snippet. The payload removes its own script tags after a few seconds, so network traces are worth more than DOM snapshots.
  3. Fetch the snippet your site actually serves. Open a page in a clean browser profile, copy the Optimizely script URL from DevTools > Network, download it and search it. Don't rely on what the dashboard shows.
  4. Search for the indicators. Look for 1688360.cc (paths /track.js, /card.php, /hit.php, /html2canvas.min.js), the .icu hosts 5vobauua3ojpklx.icu and eil786npto9u5bq.icu, and webshell.lol. Also look for unexpected atob( calls and any unfamiliar .cc or .icu domains. Get full hashes from Amirul Cyber and wallet addresses from BleepingComputer.
  5. Review Project JavaScript and every experiment's custom code for changes you didn't make, especially on November 13–16. Check the project change history for unknown users.
  6. Rotate Optimizely access. Audit users, remove stale collaborators and rotate API tokens. Report URI recommends this even though Optimizely has already reset affected accounts.
  7. Search your own logs and CSP reports. Look for requests to the attacker's domains and for sendBeacon or POST requests to unknown hosts from checkout pages.
  8. Assess what was exposed. Report URI's advice is to treat card data entered on your forms on November 16 as potentially exposed and to invalidate session tokens readable from JavaScript. Work out whether you have notification obligations, and bring in your payment processor if card data was involved.

Don't stop at Optimizely. The same actor came in through Kissmetrics, StorifyMe and Minute Media in the days before, so audit any vendor script with a dashboard-editable code field.

The controls that actually worked

Report URI's data shows which defences stopped the attack in practice. Only 24 of the affected hosts blocked the payload with their Content Security Policy. What separated them from the rest is worth understanding.

Control Did it help here? Why
CSP allowing cdn.optimizely.com No The malicious code came from the trusted host
CSP with 'strict-dynamic' No It trusts any script the allowed snippet injects
CSP with 'unsafe-eval' Partly defeated It let the inline fallback copy run even when the external file was blocked (seen at HSBC)
Restrictive script-src without the above Yes track.js from an unknown host was refused
Tight connect-src Partly It blocked exfiltration, but not the in-page address swap
Hosted payment iframes Yes, for cards The skimmer couldn't read fields inside the processor's frame
HttpOnly session cookies Yes, for sessions They're unreadable from page JavaScript
Consent manager gating (e.g. OneTrust) No It controls when a script loads, not what it contains

Practical hardening steps

  • Get experimentation and tag scripts off sensitive pages. Checkout, login, account settings and wallet pages rarely need A/B testing. Not loading a third-party script there is the cheapest control and the hardest to get wrong.
  • Write a strict CSP and run it in report-only mode first. List your real script hosts, drop 'unsafe-eval', avoid 'strict-dynamic' unless you understand what it trusts, and limit connect-src to the endpoints you actually call. Send violation reports to a collector you actually monitor. An unknown host in a script-src violation was the early warning here.
  • Know where Subresource Integrity stops working. SRI pins a file's hash, which suits static library builds. Optimizely-style snippets change whenever someone edits an experiment, so you can't hash-pin them without breaking the product. For those you need change detection: a scheduled job that fetches the served snippet, compares it with the last known-good version and alerts on any difference.
  • Use hosted payment fields from your processor rather than plain inputs on your own page.
  • Set HttpOnly and Secure on session cookies, so a page-level script has nothing worth taking.
  • Think about server-side experimentation for high-value flows. Deciding variants on the server keeps a vendor's editable JavaScript out of the browser completely.
  • Lock down vendor dashboards. Use SSO and MFA, assign named roles, run regular access reviews, and get change-history alerts wherever the vendor offers them. Anyone who can edit Project JavaScript can effectively deploy code to your production site.

If you take card payments

PCI DSS already expects this. As Report URI's summary of requirements 6.4.3 and 11.6.1 explains, payment pages must have an authorisation method for every script, a way to protect script integrity, and a written inventory with a "business or technical justification" for each script. They also need a mechanism that detects and alerts on unauthorised changes to scripts and security-relevant headers as the browser receives them. If an A/B testing snippet is on your payment page, ask what its justification is. In most cases the honest answer is to remove it.

What it means if you build a SaaS that serves customer code

The root cause wasn't JavaScript delivery. Optimizely says it was a signup-flow flaw that apparently let an outsider write into many unrelated customers' projects. Amirul Cyber also noticed the same loader line appearing under about ten unrelated organisations' project IDs, which fits a shared injection method. Neither Optimizely nor the researchers have published technical details yet, so treat any theory about the exact bug as unconfirmed.

If your product lets customers ship code, config or content into their own users' browsers (tag managers, chat widgets, feedback tools, analytics, personalisation), Optimizely's experience points to a short list of checks:

  • Test tenant isolation at the write path. Signup, invite, SSO just-in-time provisioning and account-merge flows should each have a test proving a new account can't reach an existing project.
  • Alert on bulk cross-tenant edits. One actor modifying hundreds of customer projects in a few hours should page someone.
  • Give customers an immutable change log with author, diff and timestamp, plus webhooks on code changes. Report URI could rebuild this attack only by crawling snippets from outside.
  • Give customers a kill switch and a way to restrict where custom code may run, such as excluding paths like /checkout.
  • Respond publicly and quickly. Report URI said it had not reached Optimizely through several channels by the time it published, and much of the first day's information came from third parties.

If you're choosing a vendor, these are fair questions to put in a security questionnaire. If you're listing a developer or analytics product on LaunchBoosts, clear answers to them are worth putting on your security page.

What to do now and what to watch

This week:

  • Run the eight-step check above on every site that loads Optimizely, Kissmetrics, StorifyMe or Minute Media scripts.
  • Remove experimentation and analytics tags from checkout and login pages, or justify each one in writing.
  • Deploy a report-only CSP without 'unsafe-eval' and see what breaks before you enforce it.
  • Tell staff and customers who use iPhones to update. Report URI recommends iOS 18.7.3+ or 26.3+.

Watch for:

  • Optimizely's post-incident report on its status page. In particular, watch for how long the signup flaw was exploitable and whether any data beyond injected code was accessed.
  • Notifications from affected brands about card exposure, which would show how much the skimmer actually captured.
  • Further analysis of the iOS boot.js stage, which Amirul Cyber said was still being examined.
  • Copycat attacks on other experimentation and tag platforms. This attacker had already used three other vendors before reaching Optimizely.

The broader lesson: a "trusted" domain in your CSP tells you where a script came from, not what it does. Decide which pages third-party code may run on, and keep that list short.

Frequently asked questions

What happened in the Optimizely supply-chain attack?

On November 16, 2026, an attacker used a flaw in Optimizely's Experimentation account signup flow to add malicious JavaScript to customer projects. The code was served from Optimizely's own CDN to visitors of sites including Zoom, GitLab, the BBC, CNN and Salesforce. It swapped crypto wallet addresses, skimmed card details, stole cookies and screenshots, and routed iPhone users to exploit kits.

How do I know if my site was affected?

Optimizely says it has notified all affected customers. You can also check yourself: search your Project JavaScript, experiment code and the snippet your site serves for 1688360.cc, .icu domains, webshell.lol and atob(. Then look through your CSP violation reports and project change history for November 13–16, 2026. Keep copies of any evidence before you clean anything up.

Would a Content Security Policy have stopped the Optimizely malware?

Only a strict one. Allowing cdn.optimizely.com did nothing, because the bad code came from that domain. The sites that blocked it had a script-src that rejected the attacker's domain, didn't rely on 'strict-dynamic' and didn't allow 'unsafe-eval'. A tight connect-src stopped data being sent out, but it couldn't stop wallet addresses being swapped inside the page.

I visited an affected site on November 16. What should I do?

Report URI and BleepingComputer give the same advice. Update your iPhone or iPad to the latest iOS (18.7.3+ or 26.3+ per Report URI). Check that any crypto transfers you made that day went to the address you meant. Watch your card statements for charges you don't recognise. Card numbers typed into hosted payment iframes were out of the skimmer's reach.

Is Optimizely safe to keep using?

Optimizely says it has fixed the vulnerability, removed the injected code, revoked sessions and tokens, and added protections against unauthorised code delivery. The investigation was still open on November 17, 2026. A reasonable approach is to keep using it but take it off checkout and login pages, or replace it there with server-side experimentation.

Sources

  1. Report URI: World's Largest Brands Infected With Malware Via Optimizely— blog.report-uri.com
  2. BleepingComputer: Optimizely hack infected CNN, Fox News, BBC, Salesforce sites with malware— bleepingcomputer.com
  3. Optimizely Status: Web Experimentation Security Incident— status.optimizely.com
  4. Infostealers.com: Poisoned Scripts on BBC, Zoom and Other Major Websites Linked to iPhone Exploit Loaders— infostealers.com
  5. Amirul Cyber: OPTIMJACK, the Optimizely skimmer— amirulcyber.com
  6. Report URI: PCI DSS 4.0.1 and Requirements 6.4.3 and 11.6.1— blog.report-uri.com
person

LaunchBoosts Research Desk

AI-assisted research

Explainers on software and AI industry trends, drafted with AI assistance from the public sources cited in each article and published after automated editorial checks for length, independent sourcing and originality.