Dev.to Security 🔐 Cybersecurity 👁 0 📖 8 min read

The Optimizely Web incident: what happened, and what SaaS teams should test now

This article was first published on the Rokkhe blog on Hashnode. It was drafted with AI assistance and reviewed and edited by the Rokkhe team; all facts were checked against the linked public sources as of 17 Nov 2026.

This article was first published on the Rokkhe blog on Hashnode. It was drafted with AI assistance and reviewed and edited by the Rokkhe team; all facts were checked against the linked public sources as of 17 Nov 2026.

A sourced timeline of the 16 November 2026 Optimizely Web incident, as of 17 Nov 2026. By the Rokkhe research team.

On Monday 16 November 2026, visitors to some of the busiest sites on the internet were served malicious JavaScript. The code did not come from those sites' own servers. It came from the Optimizely Web snippet those sites load for A/B testing, served from cdn.optimizely.com with a valid certificate.

This post pulls together what is publicly known as of 17 Nov 2026, how the attack worked, how Optimizely responded, and what we think SaaS teams should take from it. Several key questions are still open, and we mark them as open. All times below are UTC, as the sources report them.

What happened: the timeline

The incident was first documented by Scott Helme of Report URI, in a post titled World's Largest Brands Infected With Malware Via Optimizely. His team checked 1,611 Optimizely snippets and found malicious code in 188 of them, belonging to 178 organisations. The named list includes Zoom, the BBC, CNN, Salesforce, GitLab, IKEA, HSBC, Equifax, Air France-KLM and Ledger. Report URI notes the real number is probably higher, because it could only check the snippets it could find.

Report URI's timeline, in short:

  • 16 Nov, 02:21: last clean capture of the BBC's Optimizely snippet.
  • 04:21: first Optimizely injection seen, in the BBC snippet.
  • 05:23: the attacker's main domain is registered.
  • 08:10: first build of the payload, track.js.
  • 13:43: a card skimmer is added to the payload.
  • 17:13 to 17:57: iOS exploit delivery is added.
  • 19:24 to 19:59: Report URI's sweep finds 186 of 1,611 snippets infected.
  • About 20:40 to 21:10: mass clean-up.
  • 21:52: only 1 of 188 infected snippets still carries the code.

Report URI says the payload was rebuilt 26 times during the day. StepSecurity, working independently from public urlscan.io records, counted at least four edits in two hours. Independent researcher Amirul Cyber named the campaign OPTIMJACK, according to Cyber Kendra.

BleepingComputer independently analysed copies of modified files and used urlscan.io records to confirm the script loaded, or tried to load, on sites including Zoom, KLM, GitLab, BBC, CNN, Smartsheet and Salesforce.

Optimizely posted on its status page at 01:02 UTC on 17 November, and on X at 01:21 UTC, saying it had "identified and contained a security incident affecting some Optimizely Web Experimentation customers" and that "the vulnerable path is blocked, the injected code was removed, and customers have been contacted directly."

No threat actor has been publicly named. Optimizely had a separate, earlier incident in February 2026, a voice-phishing attack that exposed business contact information from internal systems, reported by BleepingComputer. Neither Optimizely nor any researcher has linked the two.

How the attack worked

Optimizely Web works by having each customer load a JavaScript snippet from Optimizely's CDN. The snippet carries Optimizely's own code, the customer's experiments, and an optional field called Project JavaScript: custom code the customer writes in the dashboard, which runs on every page.

According to Report URI, the attacker wrote a short loader into the Project JavaScript of many unrelated customer projects and left the rest of each snippet untouched. The loader added a script tag pointing at the attacker's server. Report URI saw three variants: a plain script tag, the same URL hidden in base64, and a full inline copy of the payload that runs through eval if the attacker's domain is blocked. Report URI is clear that this was not CDN cache poisoning. The CDN served exactly what the platform gave it.

On the entry point, Optimizely told 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." Optimizely has not yet published a technical description of that flaw, or explained how a signup flow led to write access on existing customers' projects. As iTWire put it, that identifies the affected workflow, but not the technical failure. We will not guess at the mechanism here.

What the payload did, per Report URI, BleepingComputer and StepSecurity:

  • Crypto address swapping. It found wallet addresses on the page, in the clipboard and in outgoing requests, and replaced them with the attacker's.
  • Card skimming. It watched for a card number, expiry and CVV typed into a site's own form fields and sent them out. Card fields inside a payment provider's hosted frame, such as Stripe or Adyen, were out of reach.
  • Cookie and screenshot theft. In some cases it sent the page URL, cookies readable by JavaScript (not HttpOnly ones) and a screenshot.
  • iOS exploit delivery. On iPhones and iPads it loaded the Coruna and DarkSword exploit kits in a hidden frame. Report URI says iOS 18.7.3 and later, and iOS 26.3 and later, are patched. Cyber Kendra notes that nobody has shown an iPhone was actually compromised this way.

The visible financial loss is small next to the reach. Report URI counted about $7,040 into the attacker's wallets on 16 November, StepSecurity about $6,600, and Cointelegraph about $7,100, per The Tech Boom. Card, cookie and iOS outcomes do not show on a blockchain, and nobody has said how much of that data reached the attacker.

Report URI also traced an earlier wave from 13 to 16 November, where the same actor's injector was served through other third-party scripts (Kissmetrics, StorifyMe and Minute Media). So Optimizely was one route in a wider campaign, not the only one.

How Optimizely responded

Credit where it is due: the bulk of the clean-up happened within hours of the first public report. Report URI measured infected snippets falling from 169 at 20:56 UTC to one at 21:52 UTC on the same day.

In its statement to BleepingComputer, Optimizely said it:

  • blocked access to the vulnerable path and deployed a fix;
  • removed the injected code from affected projects;
  • ended active sessions, revoked tokens and reset access for affected user accounts;
  • blocked the code from being delivered at Cloudflare;
  • notified all affected customers, and would post updates at status.optimizely.com.

That covers the right order of work: close the entry point, remove the payload, cut off any access the attacker may hold, add a delivery-layer block, and tell customers. As of 17 Nov 2026, the status page entry is still in "Monitoring" and Optimizely says its investigation is ongoing.

To the Optimizely team: thank you for the fast containment and the clear public notice. Incidents like this are hard, and the speed of the clean-up mattered. If you want to compare notes, we're happy to share our detection rules for this behaviour.

Open points, as of 17 Nov 2026:

  • How the signup flow flaw gave write access to existing projects.
  • When the attacker first got access, and how many customers were affected in total.
  • Whether any customer data held by Optimizely was accessed.

Lessons for SaaS teams

1. A trusted host is not trusted content. Allowlisting cdn.optimizely.com in a Content Security Policy did nothing here, because the malicious loader arrived from that host. Report URI crawled 256 affected hosts: 68 had no CSP at all, 50 had a CSP that let the payload and the data theft through, 17 blocked data theft but not address swapping, and 24 (15%) blocked the payload. What helped was a restrictive script-src listing only the hosts the site needs, plus a tight connect-src. 'strict-dynamic' and 'unsafe-eval' let the attack through.

2. Any field that ships code to every page is a production deploy path. Project JavaScript, tag manager containers and similar "custom code" boxes are effectively a deploy pipeline with no review. Treat write access to them like write access to your main branch: few people, strong auth, logged and alerted. It is the same idea as pinning your GitHub Actions by commit SHA: code you did not write should not change under you without anyone noticing.

3. If you run a multi-tenant platform, test the edges between tenants. The confirmed entry point was a signup flow. Signup, invite, SSO linking, account recovery and project transfer are the places where one tenant's identity can end up touching another tenant's data. They are worth testing by someone trying to break them, not just by unit tests. Our OWASP API Security Top 10 testing guide covers the object- and function-level authorization checks that matter most here.

4. Clean now does not mean clean then. As StepSecurity points out, a snippet that looks clean today proves nothing about 16 November. Keep history: CSP reports, proxy logs and your vendor's change history. Optimizely's change history records Project JavaScript edits and whether they came through the REST API or the web interface, according to Cybersecurity News.

5. If you use Optimizely Web, do the checklist. Report URI's list is a good start: check every Project JavaScript and experiment custom code for the attacker's domain, atob( and anything you did not write; review users, API tokens and integrations and rotate credentials; treat cards entered on your own forms on 16 November and secrets in non-HttpOnly cookies as exposed; deploy a CSP with reporting; and decide whether you need to tell your own customers.

Where Rokkhe helps

We will not claim we would have stopped this. Nobody outside Optimizely knows the full root cause yet. But each step of this attack maps to something that can be tested and something that can be detected, and that is the work we do.

Testing the entry point. Rokkhe's AI agents run pentests against your own platform and try to prove real exploits, not just flag possibilities. If you run a multi-tenant SaaS, that includes the flows that matter here: signup, invites, SSO linking and anything that grants a new account rights on existing data. If an agent can get from a fresh account to another tenant's project, you get the proof and the steps, not a guess.

Testing your exposure to third-party code. We check which third-party scripts run on your pages, who can change them, and whether your CSP actually blocks a loader injected from a trusted host, including the eval fallback case that got through at one affected site.

Turning proof into detection. Every exploit we prove becomes a detection rule in your SIEM. Our library has 300+ rules built from proven exploits. For an incident like this, that means rules such as: a Project JavaScript or tag container change from a new or unusual account or API token; CSP violation reports naming a script domain you have never loaded before; and outbound beacons from your pages to unknown hosts. Our SOC can watch those alerts with you, so an unexpected change to code that runs on every page gets looked at in minutes.

Keeping your data where you want it. Rokkhe can run self-hosted, in the Paris region, with your own model key, so the testing and the findings stay inside your boundary.

Book an evaluation

If you run a SaaS platform, or rely on third-party scripts that can change what your users see, we can test those paths and leave you with detections, not just a report. Book an evaluation at rokkhe.com.

Rokkhe SAS is based in Strasbourg, France. This post is based on public reporting as of 17 Nov 2026 and will be updated if Optimizely publishes more detail.

📰 Read the original article on Dev.to Security

Originally published by Dev.to Security. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.