# Optimizely attack: a marketing script on 178 websites

> On November 16, an A/B testing snippet served card-skimming malware on 178 sites, Zoom and HSBC among them. Every marketing tag runs on your checkout page too.

- URL: https://www.iqsource.ai/en/blog/optimizely-attack-marketing-scripts-178-sites/
- Published: 2026-11-20
- Author: Ricardo Argüello, IQ Source

Your marketing team decides what code runs on your checkout page. Most marketing teams don't know that.

Every tag pasted into the tag manager, the A/B testing snippet, the ad pixel, the AI chat that answers product questions, runs with the same access as the code your developers wrote. It can read whatever the visitor types. On Monday, November 16, that stopped being a hypothetical for 178 organizations.

## One login, 178 websites

The attacker never broke into any of those sites. They broke into Optimizely.

Optimizely said an unauthorized party exploited a vulnerability in its Experimentation account signup flow to impersonate a user, <a href="https://www.csoonline.com/article/4249925/optimizely-confirms-massive-cyberattack-on-customer-websites-injected-stealer-malware.html" target="_blank" rel="noopener noreferrer">as reported by CSO Online</a>. From there, per <a href="https://blog.report-uri.com/worlds-largest-brands-infected-with-malware-via-optimizely/" target="_blank" rel="noopener noreferrer">Report URI's analysis</a>, they used an optional field in each customer's snippet called "Project JavaScript" and left everything else alone. The file still came from cdn.optimizely.com, signed with Optimizely's certificate. To the visitor's browser, nothing looked wrong.

Scott Helme's team at Report URI went through more than 1,600 snippets and found 188 infected ones across 178 organizations, <a href="https://www.itpro.com/security/cyber-attacks/optimizely-hack-everything-we-know-so-far-as-hundreds-of-websites-found-serving-malware-to-visitors" target="_blank" rel="noopener noreferrer">ITPro reported</a>. Zoom, BBC, CNN, IKEA, HSBC and Equifax are in the reported list. Helme expects the real count is higher.

The payload was a grab bag. It swapped Bitcoin, Ethereum and Solana addresses for the attacker's, copied card numbers typed into the site's own forms, took cookies and screenshots, and pushed exploit kits to iPhones and iPads. Cards typed inside a Stripe or Adyen iframe were out of its reach.

<a href="https://krebsonsecurity.com/2026/11/attackers-used-optimizely-to-a-b-test-malware-on-top-websites/" target="_blank" rel="noopener noreferrer">Brian Krebs</a> notes the attackers rebuilt it 26 times during the day. They A/B tested their malware on an A/B testing platform.

## Only 24 blocked it

Optimizely blocked the vulnerable path, removed the injected code and revoked tokens. Good. That fixes Optimizely.

It does nothing for the next vendor. Your heatmap tool, your chat widget, your review badge. Any of them can be the next cdn.something.com serving someone else's code.

The number I keep coming back to is in Krebs's write-up. Of the affected sites Report URI classified, 68 had no Content Security Policy at all. A CSP is the list of domains your site accepts code from and lets data go to. Another 50 had one so loose it didn't matter. Only 24 had a policy that actually stopped the payload.

Twenty-four.

And here is the uncomfortable part for marketers. Analyst Frank Dickson told CSO that Optimizely's own documentation asks customers to allow 'unsafe-eval' in their policy. At HSBC, a backup copy of the malicious code ran through exactly that opening, even though the policy blocked the main payload. The tool's setup instructions widened the hole.

## AI widgets add tags, they don't remove them

Marketing teams are adding a lot to websites right now. AI chat, personalized recommendations, forms that fill themselves in, markup so AI assistants read the page better. Each one shows up as a snippet somebody pastes.

I'm not telling you to skip them. We run an AI chat on iqsource.ai ourselves. But each one is another copy of the key, and the person deciding to hand it out is usually the one pasting the snippet, not the one who answers for the breach.

I wrote about the engineering side of this with the <a href="/en/blog/litellm-attack-ai-trust-chain-broken/">LiteLLM attack</a> and with <a href="/en/blog/an-ai-skill-is-code-with-your-keys/">AI skills that run with your keys</a>. Optimizely is the same story, told to the team that never reads those posts.

## What I'd do this week

Open your tag manager and count. How many third-party scripts load on checkout? Who asked for each one? Which are still in use? Does any of them need to be on the page where someone types a card number?

If answering takes more than an afternoon, you've found your risk.

If you want us to look at it, we start with a page-by-page table at IQ Source: every script, who requested it, what it can read, and whether it belongs there. Then we work with your IT team on a Content Security Policy in report-only mode, so you see what would break before anything gets blocked. It sits inside our <a href="/en/services/data-driven-optimization/">data-driven optimization</a> work, because a test is only worth running on a page you can account for. Send us your checkout URL and we'll start there.

## Frequently asked questions

### What happened in the Optimizely attack of November 2026?

On November 16, 2026, an attacker exploited a vulnerability in Optimizely Experimentation and added malicious JavaScript to customer projects. According to Report URI, 178 organizations served that code from Optimizely's CDN. It skimmed payment cards, stole cookies and screenshots, and swapped cryptocurrency wallet addresses for the attacker's.

### Which websites were affected by the Optimizely attack?

Report URI found 188 infected Optimizely snippets belonging to 178 organizations. Press reports name Zoom, BBC, CNN, IKEA, HSBC, Equifax, GitLab and Starbucks, among others. Salesforce said it was investigating unauthorized modifications to Optimizely scripts and disabled them on its sites.

### Would a Content Security Policy have stopped the Optimizely attack?

For some sites it did. Report URI found that 24 of the affected sites it classified had a Content Security Policy that fully blocked the malicious Optimizely code. Another 50 had a policy too permissive to help, and 68 had none. A strict policy limits where a compromised script can send stolen data.

### How can a marketing team reduce the risk of third-party scripts like Optimizely?

A marketing team reduces the risk of third-party scripts like Optimizely by keeping an inventory of every script on every page, removing the ones that are not needed on checkout and payment forms, and agreeing with IT on a Content Security Policy that limits which domains can load code and receive data.
