Cyberattack Case Study: How One Vendor Flaw Put Malicious Code on 178 Websites

On November 16, 2026, a cyberattack on one website vendor sent malicious code to visitors of 178 organizations at once, including Zoom, the BBC, CNN, IKEA and Air France-KLM. The attackers did not break into any of those companies. They exploited a flaw in Optimizely, a tool the sites used to test page designs, and their code rode along on every page that loaded it. This case study walks through what happened, who responded, what is still unknown, and what other organizations can learn.
Key Takeaways
- One vendor, 178 organizations. Report URI found 188 infected Optimizely scripts across 178 organizations, including 32 finance or banking sites.
- The code was built to steal. It swapped crypto wallet addresses, skimmed card numbers, stole cookies and screenshots, and pushed older iPhones and iPads to exploit kits.
- Proven losses are small so far, but unknown losses are not. About $7,000 in cryptocurrency was traced. Nobody outside the attackers knows how many cards, sessions or devices were taken.
- No AI involvement was found. The researcher who uncovered the attack said there was no evidence of it.
- Third-party scripts are part of your attack surface. Content Security Policy, script monitoring and limits on who can edit vendor code are the main defenses.
The Cyberattack at a Glance
Diagram
The Cyberattack at a Glance
- When
- What
- How
- Who found it
- Scale
- Exposure
- Known theft
- AI involvement
| Question | Answer |
|---|---|
| When | November 16, 2026, with earlier activity seen in September |
| What | Malicious JavaScript added to customer projects in Optimizely Web Experimentation |
| How | A flaw in Optimizely's Experimentation account signup process that let attackers impersonate customer administrators |
| Who found it | Scott Helme and the team at Report URI |
| Scale | 188 infected scripts out of 1,611 checked, across 178 organizations |
| Exposure | About 34 million page visits loaded the code (range 6 million to 150 million) |
| Known theft | About $7,000 in cryptocurrency |
| AI involvement | None found |
Timeline of the Attack
- September 2026. Report URI first saw the same injected code on Turkish websites, The Record reported.
- November 16. Malicious code appeared in Optimizely projects used by major sites. Report URI raised the alarm, and Optimizely said it had "identified and contained a security incident affecting some Optimizely Web Experimentation customers."
- November 17. Optimizely confirmed in a technical advisory that a vulnerability let attackers inject JavaScript. Salesforce disabled Optimizely scripts, Equifax stopped using the tool, and IKEA began investigating.
- November 17 to 18. More companies reviewed their exposure. HSBC said unauthorized scripts were not active on its websites, and Apple said it had already patched the flaws the iPhone exploit kits targeted, Reuters reported.
Most of the infected files were cleaned within a day of the report, according to Helme. Check the Stack Row cyberattack tracker for the latest status of each named organization.
How the Attack Worked
Optimizely lets marketing and product teams change a live website without a code release. A small script on each page loads the customer's "project," which can include custom JavaScript. That flexibility is what the attackers used. Optimizely told BleepingComputer the vulnerable workflow was its Experimentation account signup process, which let the attacker act as customer administrators without their passwords and change the code their sites served.
[[flow: How the malicious code reached visitors | Attackers abuse a flaw in Optimizely's account signup process > They add a loader to customers' Project JavaScript > Every site that loads the project runs the loader > The loader fetches the attack code from the attackers' servers > The code runs in each visitor's browser]]
Two design choices made the attack worse. First, A/B testing scripts are placed near the top of the page so they can change content before it appears. Second, they usually run before any cookie consent prompt. Helme told Krebs on Security that the code is set up to fire immediately on page load, so the attack did not wait for visitors to agree to anything.
Once loaded, the code tried several things:
- Crypto clipping. Replaced wallet addresses that visitors copied with the attackers' addresses.
- Card skimming. Read card numbers typed into forms the sites built themselves.
- Cookie and screenshot theft. Grabbed session cookies that scripts can read, which could let attackers take over logged-in sessions.
- iOS exploits. Sent iPhone and iPad visitors to the Coruna and DarkSword exploit kits, which can take over outdated devices without a click.
Payment forms hosted in a payment provider's own frame were out of reach, because a script on the main page cannot read inside another site's frame.
Who Was Affected
Report URI named organizations across technology, media, finance, travel and retail. They include Zoom, the BBC, CNN, Salesforce, GitLab, IKEA, HSBC (UAE), Equifax, Air France-KLM, Ledger, Peacock, Fox News and Alaska Airlines. Thirty-two were finance or banking sites.
Being on the list means a company's website loaded an infected script. It does not mean the company's own servers or customer databases were breached. As one researcher put it to Newsweek, the sites' own code and infrastructure were fine; the trusted vendor script was the problem. Report URI also said the true number of affected organizations is likely higher, because it could only check the scripts it found in time.
How the Companies Responded
| Organization | Response |
|---|---|
| Optimizely | Blocked the vulnerable path, removed the injected code, ended sessions, revoked tokens and contacted customers |
| Salesforce | Disabled Optimizely scripts on preview and production sites; will notify customers if data was accessed |
| Equifax | Stopped using Optimizely; found no evidence of compromise to its systems or data |
| IKEA | Investigating any impact on its digital services and customers |
| HSBC | Said unauthorized scripts were not active on its websites |
| Apple | Said it had already patched the iOS flaws the exploit kits targeted, so up-to-date devices were protected |
The pattern is clear: customers moved fast to switch the vendor off, then investigated. That is the right order for a supply-chain incident, because the vendor's fix does not undo what already ran in visitors' browsers.
What Is Still Unknown
- Card losses. Researchers can trace crypto on public blockchains, but not skimmed cards.
- Session theft. Stolen cookies leave no public trace, and their value depends on each site's security settings.
- Device infections. Researchers have not confirmed a successful infection. Apple says up-to-date devices were protected, but older ones may not have been.
- The full victim list. Some infected scripts may have been cleaned before researchers saw them.
- The attackers. Helme said the same wallet had been receiving funds since March 2026, and the same infrastructure was injected through at least three other script vendors in the days before. That points to an established criminal operation, not a first-timer.
Was AI Involved in This Cyberattack?
No evidence shows it. Helme told Newsweek that researchers saw "no evidence of AI involvement." One threat researcher said he would not be surprised if offensive AI tools were involved, but that was speculation, not a finding.
That matters because AI is now the default suspect in big incidents. Stack Row's report on rogue AI agents found the same thing across recent outages and attacks: the causes were vendor flaws, bugs, misconfigurations and human criminals.
The Same Week: Other Cyberattacks and Outages
The Optimizely attack landed in a crowded week:
- November 19: A ransomware group leaked data on thousands of Visa, Mastercard and JPMorgan Chase customers after hitting an unnamed third-party provider in the Asia-Pacific region, according to The Wall Street Journal. Visa said no card numbers were affected.
- November 17: Apple services went down after an internal update, not an attack.
- November 16: Chase and other US banks had outages that Chase blamed on a technical bug.
- November 11: United Airlines blocked third-party AI agents from booking flights, citing security risks from shared credentials.
Two of these were real cyberattacks, and both came through third parties. That is the common thread: attackers increasingly go after the vendor that many companies trust, rather than each company one by one.
How to Defend Against Third-Party Script Attacks
No single control stops this kind of cyberattack, but layered ones limit the damage.
- Inventory every third-party script. List what loads on each page, who owns it and why it is there.
- Use a strict Content Security Policy. Allow scripts only from domains you trust, and avoid blanket allowances. The OWASP Content Security Policy Cheat Sheet explains the options.
- Monitor in real time. CSP reporting and script-change monitoring flag new domains and changed code as soon as they appear.
- Limit who can edit vendor-hosted code. Treat the custom-code field in a marketing tool like production code, with strong sign-in, multifactor authentication and change review.
- Keep payments in hosted frames. Card fields hosted by a payment provider are out of reach of page scripts.
- Mark session cookies HttpOnly. Scripts cannot read cookies marked HttpOnly.
- Have a kill switch. Know how to remove a vendor script from every page in minutes.
For context on how often these incidents happen, see Stack Row's cybersecurity statistics and data breach statistics.
What Visitors Should Do
If you visited an affected site on November 16, 2026:
- Log out of all sessions on the sites you used, and change those passwords.
- Check card statements and crypto transfers, and enable transaction alerts.
- Turn on multifactor authentication where you can.
- Update your iPhone or iPad to the latest operating system.
Simply opening a page does not mean information was stolen, but these steps close the most likely gaps.
Frequently Asked Questions
What Is a Supply-Chain Cyberattack?
A supply-chain cyberattack compromises a supplier so that its customers are affected. Here, attackers changed scripts from a vendor that many websites trusted, so the malicious code ran on all of them.
Which Websites Were Affected by the Optimizely Attack?
Report URI named 178 organizations, including Zoom, the BBC, CNN, Salesforce, GitLab, IKEA, HSBC (UAE), Equifax, Air France-KLM, Ledger, Peacock, Fox News and Alaska Airlines. Being named does not mean the company itself was breached.
Is the Optimizely Attack Over?
Optimizely says it fixed the signup flaw and removed the injected code, and researchers said infected files were cleaned within about a day. Investigations into what was stolen are continuing.
How Much Money Was Stolen?
Researchers traced about $7,000 in cryptocurrency. Losses from skimmed cards, stolen sessions and infected devices are unknown.
Could a Content Security Policy Have Stopped It?
A strict policy limits which domains scripts can load from and can report suspicious activity. It would not block code served from an already trusted vendor domain, but it can stop that code from fetching more code or sending data to the attackers' servers.
Was This Cyberattack Carried Out by AI?
No evidence shows that. The researcher who found it said there was no sign of AI involvement.
Resources
- Black Kite: Optimizely supply-chain attack hits 178 organizations
- FinChannel: Optimizely fixes flaw after websites served malicious code
- The Record: Optimizely confirms hackers put malicious code into customer sites
- Krebs on Security: Attackers used Optimizely to A/B test malware on top websites
- Newsweek: Cyber attack hits 178 major websites, full list
- OWASP: Content Security Policy Cheat Sheet
- Investing.com via Yahoo: Visa, Mastercard and JPMorgan customer data leaked
