Your A/B testing vendor just executed malware in the browsers of CNN, BBC, and Zoom visitors.
This is a cybersecurity problem your firewall never saw. Optimizely Web Experimentation is the kind of script teams wave through because marketing owns it and the domain looks legitimate. Attackers exploited a flaw in that service, planted malicious code in scripts customers load on every page, and let those sites do the distribution. Report URI says visitors to at least 178 organizations got the payload, including GitLab, Salesforce, Equifax, HSBC, and IKEA.
They shipped a dropper on a button-color test.
The Experiment Script Became Payload
The code rode in on a script you already trusted, on pages you already published, in front of users who already clicked through your cookie banner.
Optimizely said an attacker abused a flaw in Web Experimentation and planted malicious code in some of the scripts customer sites load from the service. Reporting from The Hacker News, citing Report URI telemetry, says that payload was built to drop a card skimmer and iOS exploit kits. Checkout fields and mobile browsers were the targets.
Your brand was the CDN.

This is Magecart tradecraft with a SaaS badge. The supply chain sat inside an experimentation platform that thousands of sites include by design, usually on the hottest paths: home, product, pricing, signup. SecurityWeek’s roster reads like a board-deck customer list. This is a bad look for every security leader who still lets marketing own tags with no hash and no kill switch.
Your threat-protection stack probably logged nothing useful. The request went to a known hostname. TLS looked fine. Content Security Policy, if you had one, likely allowed that origin because the business demanded experiments on the live site. Threat detection that still spends its week on brute-force noise against SSH and VPN will not flag a first-party page loading its approved vendor.
SOC playbooks still grade themselves on server patches and identity alerts. A signed origin with a poisoned file fails none of those checks. The user paid. The skimmer collected. Your dashboard stayed green.
Uber’s letters the same week are the other flavor of the same disease. A third-party provider told the company on October 9 that stolen internal credentials were used to reach Uber’s account. Nearly 150,000 people are in those notices. Different blast radius. Same pattern: you outsourced a function and inherited an incident.
Perimeter Cybersecurity Never Sees Checkout
The browser is an execution environment you do not patch and do not inventory with the same seriousness you give servers.
Defense in depth that ends at the WAF leaves card data and mobile visitors exposed the moment a blessed origin turns hostile. Security hardening on your origin does not hash the JavaScript a vendor CDN will serve tomorrow morning. Cyber security programs still treat tag managers and experimentation pixels as marketing plumbing. Attackers treat them as remote code execution with a corporate badge.

Incident response for this class of event starts in the browser. You need the script URL, the hash that changed, the window when CSP or Report URI alerts fired, and every template that included the tag. If your IR runbook opens with firewall denies, you will spend the first hour collecting empty packets while finance asks why the payment page was still live.
iOS exploit kits in a marketing script should also kill the comfortable story that mobile users skip the worst of the web. They load your pages. Your pages load Optimizely. That is the chain.
Allow Lists Beat After Action
Stop treating vendor JavaScript as decoration. Treat it as production code with a kill switch.
Do this now, then keep doing the unglamorous parts:
- Inventory every third-party script on production templates, tag managers, and experimentation snippets. Include analytics, chat, A/B tests, and anything that can see a payment form.
- Pull experimentation tags off checkout and login until the vendor’s integrity story is in writing. Marketing can live without a headline test for a day.
- Turn on CSP with a tight script-src list and a report endpoint you actually read. If a host must stay, pin it and watch for new URLs under that origin.
- Where the vendor supports Subresource Integrity, require hashes in the template. If they rotate files without hashes, treat that as a named risk acceptance with an owner.
- Hunt the client side: new domains in waterfalls, form listeners on PAN fields, unexpected eval, and script hash changes against yesterday’s baseline.
- Put vendor JS in change control. New tags need the same review as a firewall rule. Contract for a security contact, a kill switch, and a notification SLA measured in hours.
Keep a daily diff of script hashes on the templates that handle money and credentials. That is cheaper than explaining a week of skimmer dwell to a card brand.
If you cannot name the scripts on your payment page from memory, you already know who served the skimmer.
Sources
- Optimizely Vulnerability Exploited to Serve Malware on Major Websites
- Hijacked Optimizely Scripts Served Card Skimmer and iOS Exploit Kits to CNN, BBC and Zoom Visitors
- Uber says nearly 150,000 impacted by third-party breach
Take Control of Your Server Security
Don't let brute-force attacks slow you down. Try IPBan Pro risk-free for 30 days.
Secure. Automated. Lightweight.
Take Control of Your Server Security
Don't let brute-force attacks slow you down. Try IPBan Pro risk-free for 30 days.
Secure. Automated. Lightweight.
