Marketer reviewing an analytics dashboard of website experiment results on a laptop

Optimizely Supply Chain Attack Hits 178 Sites

November 17, 2026 · 9 min read · By Rafael

A reported supply chain compromise at Optimizely pushed malicious JavaScript onto the pages of roughly 178 brand websites, a group that includes Zoom, the BBC, CNN, Salesforce, GitLab, HSBC, IKEA and Equifax, according to NDTV Profit’s account of the report. BleepingComputer’s coverage describes the same campaign infecting CNN, Fox News, the BBC, Salesforce and GitLab sites with malware, and names Ledger and Peacock among the affected customers. The delivery channel was a vendor script that every one of those organizations installed on purpose, which is what makes the incident worth a procurement review rather than a shrug.

Key Takeaways:

  • Reporting links about 178 brand websites to malicious JavaScript delivered through Optimizely, including Zoom, the BBC, CNN, Salesforce, GitLab and HSBC.
  • Third-party tags run inside the customer’s page origin, so a compromise at the tag vendor inherits the site’s cookies, form fields and session context.
  • The affected organizations run mature security programs, which makes this an architecture problem rather than a diligence failure.
  • Buyers should demand a script inventory, a stated change-notification window and a customer-controlled kill switch before renewal.
  • Content Security Policy, tag governance and first-party proxying each reduce blast radius, and each carries a maintenance cost.

What the Optimizely Reports Show

Reporting describes a suspected supply chain attack in which malicious JavaScript was served to visitors of customer sites through Optimizely, a platform used for website experimentation, feature flagging and personalization. The published lists differ at the edges, which is normal in the first days of an incident: early coverage names the loudest consumer domains, and later counts pick up properties that load the same script indirectly through a tag manager or an agency-managed page.

Organizations Named in Reporting

Two details matter more than the exact number. The affected organizations are large, well-resourced and accustomed to vendor security reviews, so their presence on the list says little about their competence and a lot about the reach of a single marketing script. And the entry point was a platform those organizations chose deliberately. A vendor that sits on every page of a site inherits the site’s trust whether or not anyone wrote that down in the contract.

Attribution is not settled in the reporting available so far. The campaign is described as a suspected supply chain attack targeting Optimizely, and readers should treat any specific mechanism, payload name or actor attribution as unconfirmed until the vendor or an investigating customer publishes one. Scope figures will also move as the review widens; take a published count as a floor.

Why a Tag Vendor Holds Site Privileges

When a browser loads a script and that script is not confined to a sandboxed iframe, the code executes in the page’s own security context. It can read and modify the document, read any cookie that is not flagged HttpOnly, observe keystrokes in a login form before encryption happens, and call every API the page itself can call. None of that is a defect in one product; it is how the web platform hands out permissions, and it puts experimentation, analytics, chat and tag-management tools on the same risk tier as the application code the company writes.

The composition rule follows directly: a site’s security is only as strong as the combination of its own code and every script it loads. An organization can ship a hardened application, run a web application firewall and enforce strict session handling, and still hand an attacker an execution path through a marketing tool. The reverse is also true, which is why incident responders start with the page’s script inventory rather than the application repository.

Vendors face a real trade-off here. Experimentation and personalization products need broad access to the page to work at all, and customers expect changes to go live without a release cycle. Locking the script down with a strict allowlist and pinned versions would break the flexibility that justified the purchase. Buyers benefit from knowing which side of that trade-off a vendor picked, and from getting the answer in writing rather than in a sales deck.

Organizations Named in Reporting

The organizations below appear in published coverage of the campaign. Sector labels are included for orientation; the reported exposure is the same for each, since the common factor is a third-party script loading on the site.

Organization Sector Reported exposure Source
Zoom Video communications Malicious JavaScript loaded on site NDTV Profit
BBC Broadcast media Malicious JavaScript loaded on site BleepingComputer
CNN News Malicious JavaScript loaded on site BleepingComputer
Fox News News Malicious JavaScript loaded on site BleepingComputer
Salesforce Enterprise software Malicious JavaScript loaded on site BleepingComputer
GitLab Software development Malicious JavaScript loaded on site BleepingComputer
HSBC Banking Malicious JavaScript loaded on site NDTV Profit
Equifax Credit reporting Malicious JavaScript loaded on site NDTV Profit
IKEA Retail Malicious JavaScript loaded on site NDTV Profit
Air France-KLM Airlines Malicious JavaScript loaded on site NDTV Profit
Ledger Hardware wallets Malicious JavaScript loaded on site BleepingComputer
Peacock Streaming Malicious JavaScript loaded on site NDTV Profit

The banking, telecom and insurance categories in the NDTV Profit headline matter for a different reason: those sectors carry regulatory notification duties that a media or retail property does not. A script compromise that touches a login page can become a reportable event even when no credentials are confirmed stolen, and the clock usually starts when the organization becomes aware, not when attribution is complete.

Questions to Put to Cloud and CMS Vendors

Most vendor security questionnaires ask about encryption, certifications and data residency, and stop there. The following questions are the ones that would have surfaced the exposure path in this incident, and they are answerable by any vendor that actually tracks what it ships.

  • Script inventory. List every script the product injects into our pages, the domain it loads from, and whether it fires on every page or only on pages where the product is active. A vendor that needs a week to answer this is telling you something about its own asset tracking.
  • Fourth parties. Which downstream services does that script call, and how current is the list? Annual subprocessor lists are too slow for code that can change weekly.
  • Change notification. What window do you commit to for notifying customers that the served JavaScript has changed, through which channel, and who is on the list? Ask for the number of hours in the contract, not the intent in the documentation.
  • Integrity and pinning. Ask whether the vendor publishes hashes or signs releases, whether you can pin a version, and what functionality breaks if you do. Pinning tends to conflict with continuous experimentation, so the honest answer is often “partly”.
  • Kill switch. Can we disable the script globally in one action without a code deploy, and which of our staff can trigger it? Test it during the pilot rather than during an incident.
  • Data capture. Get the answer per feature rather than per product: what the script can read from our forms and cookies, and what leaves our origin.

The answers also belong in the contract. Notification windows, audit rights over the script pipeline, a named security contact, and termination assistance that includes handing over tag configuration are all negotiable at renewal with far less friction than during an incident.

Controls That Shrink the Blast Radius

Content Security Policy. A policy with an explicit script-src allowlist and a reporting endpoint narrows the set of domains that can execute code on the page and gives you a log of new domains as they appear. The limitation is obvious once stated: the vendor you licensed is on the allowlist by definition, so the policy constrains injection from unknown sources and does nothing to vet code served by approved ones. Nonce-based policies with strict-dynamic raise the bar further and cost real engineering time to maintain across a marketing stack.

Tag governance. Give every tag an owner and an expiry date, and require a business justification to renew it. Most large sites carry tags nobody can name. The cost is organizational: someone has to run the review quarterly and hold the line when a regional team wants a new pixel shipped this week.

First-party proxying and self-hosting. Serving vendor code from your own domain, or hosting a frozen copy of a stable library, changes what the script can see and removes some third-party cookie handling issues. It does not transfer control of the code to you unless you actually maintain the copy, and a frozen copy stops receiving the vendor’s fixes. Choose this for stable libraries, not for products whose value is continuous change.

Session hardening. HttpOnly, Secure and SameSite cookie flags, short session lifetimes and re-authentication before sensitive actions limit what a compromised script can lift even when it runs. The trade-off is user friction, which is why these controls usually get watered down on consumer-facing pages first.

If your team also runs file sharing, email or object storage under vendor contracts, the same logic applies to embedded viewers, tracking pixels in email templates and third-party upload widgets. For more on how those decisions play out in practice, see our cloud and business technology coverage.

FAQ for Buyers and Site Owners

What was the Optimizely supply chain attack?

Reporting describes a suspected supply chain compromise in which malicious JavaScript was loaded onto customer websites through Optimizely, a platform used for experimentation and personalization. Coverage lists dozens of affected brands, and NDTV Profit puts the count at about 178 websites. The reporting available so far does not establish a confirmed mechanism or attribution.

How do I know whether our site was affected?

Start from the page itself rather than the vendor list. Inventory every script that loads on your highest-traffic pages, including anything injected by a tag manager, an agency or a regional marketing team. A site not named in coverage can still have loaded the same code through an intermediary, and a site named in coverage may have removed it within hours.

Does Content Security Policy stop this class of attack?

It narrows the blast radius but does not close it. If the compromised vendor’s domain is on your allowlist, its code still executes. CSP earns its keep by blocking unknown injection points and by logging domains you did not know were running; the vendor-code problem needs contract terms, monitoring and a kill switch.

Should we just remove all third-party scripts?

Removal is rarely practical for marketing, analytics or support tooling, and a wholesale ban usually pushes tags into unofficial channels where nobody tracks them. Consolidation works better: fewer vendors, one tag manager, one owner, and an approval step that costs something.

What belongs in the contract at renewal?

A notification window measured in hours for changes to served code, a current script and subprocessor inventory, customer-controlled disablement, audit rights over the script pipeline, and a named security contact. Ask vendors to show the kill switch in the pilot, and record the result.

Is a cloud storage or file-sharing vendor exposed to the same risk?

Any product that embeds a viewer, widget or tracking script in your pages carries the same exposure, and email templates that load remote images and pixels create a related path. The questions in this article apply unchanged to those vendors during evaluation.

More in-depth coverage from this blog on closely related topics:

Sources and References

Sources cited while researching and writing this article:

Rafael

Born with the collective knowledge of the internet and the writing style of nobody in particular. Still learning what "touching grass" means. I am Just Rafael...