Sentrix

Guide · Third-party risk

Script vendors: what to ask them

Optimizely, Polyfill.io, DebugBear: a compromised script vendor runs its code at your customers' end. The questions to ask it before the next incident.

By Sentrix · Published 2026-11-17

On 16 November 2026, according to Report URI, malicious code was served through Optimizely to visitors of 178 organisations. It was not a first. In 2024, the bought-out Polyfill.io domain served conditionally activated malicious code while more than 100,000 sites used it, according to Sansec. On 29 October 2026, malicious JavaScript served from a DebugBear address that Polymarket's site loaded dumped or moved traders' positions, according to SecurityWeek. And Hudson Rock ties the Optimizely incident to poisoned scripts at Kissmetrics, StorifyMe and Minute Media from 13 November. The full story is in our analysis of the Optimizely incident.

The common thread: a script vendor does not need to hold your data to expose your customers. Its code runs at their end, on your pages. Here are the questions to ask it.

Who can change the code served to your visitors?

At Optimizely, Project JavaScript runs on every page where the snippet loads. According to the roles documentation, only administrators and project owners can change snippet settings, and several roles can publish. The security documentation recommends requiring two-step verification for collaborators and offers SAML single sign-on. It acknowledges that the administrator and project owner roles "present a greater security risk".

Ask. Who on your side holds those roles today? Is two-step verification or single sign-on mandatory? Which API tokens and integrations can write code? And on the vendor's side: which employees or systems can change the content served to your visitors?

On 16 November, that answer was a blind spot. Optimizely places the flaw in the signup flow and has not explained how it gave write access to other customers' projects. Access controls on your accounts do not protect against a flaw in the platform itself.

How will you know the code changed?

Optimizely's documentation says the change history records every change to Project JavaScript, with author and time, and separates the interface from the API. It states no retention period.

Ask. How long is the history kept? Can it be exported to your logging? Would a change made by the attacker through a platform flaw appear in it? And on your side, who monitors the snippet as received by the browser, and how often?

How long before you are told?

Report URI saw the first injection at 04:21 UTC on 16 November. According to its own post, it tried to reach Optimizely and had no response at the time it published. Optimizely's confirmation came on 17 November at 01:02 UTC, after the clean-up.

Ask. What notice period does the contract set after a compromise is discovered? Through which channel, to whom on your side? Will the notice list your affected projects, the times and the injected content? Those are the facts you will need to assess the incident under Law 25 and PIPEDA, as we explain in the article on your duties.

Where does its code run?

A JavaScript snippet can run beyond the main site: regional sites, subdomains, a snippet served from your own domain, a mobile app that displays web pages, a smart TV app built in HTML and JavaScript. Report URI did find affected sites serving the snippet from their own domain.

Ask. On which surfaces does the vendor's code run, and which ones does your inventory cover? If only one team knows that the TV app loads the same snippet as the website, your incident assessment will be incomplete.

What does the contract say?

In Québec, the Act respecting the protection of personal information in the private sector requires a written contract, with the protection measures set out, to entrust personal information to a service provider. Whether that requirement covers a script vendor depends on the facts. Its code has the technical ability to read your forms, but ability, actual access and communication of information to a provider are separate questions, to analyse with your advisers.

What we recommend adding to the contract. The security measures of the publishing platform, prompt incident notice, access to change logs, cooperation with the investigation and the right to suspend the script without penalty.

Our view at Sentrix

  1. Rank script vendors by what they can run, not by what they store. An experimentation tool on the login page is a critical vendor, even if it keeps no data. Our third-party risk module lets you rank vendors by criticality and type of data access.
  2. Ask before the incident. Our vendor questionnaires cover access and notification. Add the questions about changing the served code and the surfaces it runs on.
  3. Check what the code does on your side. The vendor's answer does not replace your own Content Security Policy and monitoring, covered in our article on what CSP blocked.
  4. Know how to cut it off. Every third-party script should be removable in minutes, by a named person, without waiting for a release.

The next step

This week, list your script vendors and note, for each, who can change the served code, how you would be told and where that code runs. The empty cells are your priority. If you want to run this exercise with us, contact us.

Sources

Frequently asked questions

Q01
Why treat a script vendor as a critical vendor?
Because its code runs in your customers' browsers, on your pages, with access to what they see and type. On 16 November 2026, an attacker wrote into the Project JavaScript of Optimizely projects, and that code was served to visitors of 178 organisations according to Report URI. A vendor that stores none of your data can still expose your customers.
Q02
Which questions come first?
Three questions first. Who, on your side and on the vendor's, can change the code served to your visitors, and with what authentication? How soon after a compromise will you be told, and with what evidence? On which surfaces does its code run: website, mobile app, smart TV app? The answers tell you whether you will be able to assess an incident.
Q03
Is a questionnaire enough?
No. A questionnaire tells you what the vendor claims. You also need to check what its code does on your side: an inventory of the scripts actually loaded, a Content Security Policy, violation reports and change monitoring. The contract should also provide for incident notice and access to logs, beyond the written contract that Law 25 requires for any provider handling personal information.

Let's talk about your compliance program.

Last updated: