Sentrix

Analysis · Supply chain

Optimizely: a trusted dependency turned attack vector

According to Report URI, Optimizely served malicious code to visitors of 178 organisations on 16 November 2026, 8 of them tied to Canada. What remains open.

By Sentrix · Published 2026-11-17

On 16 November 2026, according to Report URI's investigation, malicious JavaScript was served to visitors of 178 organisations through the Optimizely Web experimentation platform. That is the number of organisations tied to the infected snippets the researchers found. None of these organisations needed to be breached. The code arrived inside a file they already loaded, from a domain they had approved, with a valid certificate. The incident is more than card skimming. It shows what an organisation hands to a vendor when it lets that vendor run code in its customers' browsers, and how hard it is afterwards to measure what happened.

In this article we separate what the sources establish, what researchers observed, what organisations stated, what follows technically, what remains a hypothesis and what is unknown.

An attack delivered by legitimate scripts

Every Optimizely Web customer places a JavaScript snippet in its pages, served by Optimizely's CDN. The snippet holds the platform's code, the customer's experiments and an optional field, Project JavaScript, which Optimizely's documentation calls "the first point of execution in the snippet". According to the investigation by Report URI, the attacker wrote a loader into that field that pulled in an outside script, track.js. The rest of the snippet was untouched. Sites that serve the snippet from their own domain were affected too, since the content comes from the same project.

What the incident is not. Report URI states that this was not CDN cache poisoning: the CDN served exactly what it was given. Nor is it proof of an intrusion into the internal systems of the organisations on the list.

What Optimizely stated. On its status page, the company writes that an unauthorised party "exploited a vulnerability in Experimentation and used it to add malicious JavaScript to some customer projects", and that it fixed the flaw, removed the code and secured the affected accounts. According to StepSecurity, a spokesperson placed the flaw in the account signup flow, and the company says it ended active sessions, revoked tokens and notified affected customers. How a signup flaw gave write access to the projects of unrelated customers is still unexplained. Report URI: "That's for Optimizely to explain."

A valid TLS certificate protects the connection and authenticates the server. It says nothing about what the file contains. A trusted domain can serve hostile code if its publishing path is hijacked. That is the core mechanism here: the abuse of a trust relationship granted to third-party code.

An evolving, multi-purpose payload

Report URI describes a single payload, track.js, whose functions were active at the same time. What a visitor experienced depended on their device and on what they did on the page. Not every visitor saw every behaviour.

Cryptocurrency address swapping. The code replaced valid wallet addresses in page text, forms, links, the clipboard and some requests. Transfers to the attacker's wallets were observed.

Card skimming. Added at 13:43 UTC, it read the number, expiry date and security code typed into the site's own forms. It could not directly read the fields of a payment frame hosted on another origin, such as Stripe's or Adyen's.

Cookies, page address and a screenshot. These were sent after an address swap next to payment wording. StepSecurity notes that cookies marked HttpOnly were out of reach. It would therefore be wrong to write that every cookie of every visitor was stolen.

iOS exploit kits. On iPhones and iPads, the code loaded the Coruna chain, for iOS 13 to 17, and DarkSword, for iOS 18. Delivering a payload, attempting an exploit, succeeding and suffering the consequences are 4 different things. The sources document delivery, not the number of compromised devices.

Timeline, in UTC and Québec time

UTCQuébecEvent, per Report URI unless noted
13 to 16 Nov.Earlier wave through Kissmetrics, StorifyMe and Minute Media scripts
16 Nov., 04:2115 Nov., 23:21First injection seen in an Optimizely snippet, the BBC's
08:1003:10First build of track.js, the address swapper
13:4308:43Card skimmer added
17:13 to 17:5712:13 to 12:572nd iOS host and the Coruna frame added
19:24 to 19:5914:24 to 14:59Sweep: 186 infected snippets out of 1,611
About 20:40 to 21:10About 15:40 to 16:10Mass clean-up
23:1118:11All observed files clean
17 Nov., 01:0216 Nov., 20:02Optimizely confirms on its status page

The payload was rebuilt 26 times during the day. Report URI saw 3 forms of loader: a plain script tag, the same address hidden in base64, and a full inline copy that runs if the attacker's domain is blocked. A visitor at 9:00 Québec time and a visitor at 13:00 did not receive the same code.

Why 178 does not tell the whole story

The sample is incomplete. Report URI examined 1,611 snippets found mostly in its own crawl data and found 188 infected, tied to 178 organisations. It writes: "The real number is likely higher as we could only check snippets we could find in time." Optimizely speaks of "some customer projects", with no number.

Organisations are not visitors. One organisation can serve millions of pages. Conversely, being in the inventory does not mean every visitor was exposed. Some of the 178 entries are also brands, subsidiaries or regions of one group: HSBC, for example, appears as 11 regional entries.

The payload changed over time. Smartsheet's site was cleaned, then re-infected with the hidden version: the attacker kept access and edited projects one by one. Hudson Rock stresses that different hashes of the BBC file are "different observed versions". The capabilities of one version do not carry over to the others, and the relative harmlessness of one version proves nothing about the rest. Report URI writes that every visitor received the same script: that holds for the versions it captured. Nothing published establishes that no other payload ran, or the opposite.

Controls differed from site to site. According to Report URI, Content Security Policy blocked the code at 24 of the 159 organisations it assessed, and blocked the data theft at 17 more. For each site, one has to separate modified script, loaded payload, executed payload and blocked behaviour. We cover those results in our article on what CSP blocked.

Exfiltration is not measured. The only public amounts are wallet receipts, about US$7,000 according to Report URI. It adds: "We have no way to see how many credit cards were skimmed, how many cookies or screenshots were stolen, or how many iOS devices were infected." That lack of measurement shows neither massive exfiltration nor its absence.

Our conclusion: the extent of visitor exposure remains insufficiently quantified in public information. That does not allow us to claim the real impact exceeds the published numbers.

Canada in the inventory

OrganisationDomains per Report URILink to Canada, per the source
Equifax Canadamy.equifax.ca, www.equifax.ca.ca domains
Cravewww.crave.ca.ca domain
CTV Newswww.ctvnews.ca.ca domain
BNN Bloomberg and CP24www.bnnbloomberg.ca, www.cp24.com.ca domain (bnnbloomberg.ca)
TSNwww.tsn.ca.ca domain
Hootsuitehootsuite.com, www.hootsuite.comCompany headquartered in Vancouver
Intelexwww.intelex.comCompany headquartered in Toronto
Starbucks Canadawww.starbucks.ca.ca domain

Statements. As of 17 November we found no statement specific to Equifax Canada, nor from Crave, CTV News, BNN Bloomberg, CP24, TSN, Hootsuite, Intelex or Starbucks Canada. The absence of a public notice the day after an incident is no basis for presuming a failure.

The company's systems or the visitors' browsers: a key difference

According to a review published by HackerNoon, Equifax told Reuters it stopped using Optimizely and found no evidence its own systems were compromised. That statement may be accurate. It still does not answer the question a visitor asks.

Where the attack ran. The malicious code did not need to get into Equifax's servers. It arrived inside a third-party script the site loaded, and it could run in the browser of a visitor who received that version, depending on the page, the time and the controls in place. There, it could see what the page displayed, read what the visitor typed into a form on the site before the information even left for the servers, and send it elsewhere.

Why both claims can be true at once. Intact servers and exposed visitors are not mutually exclusive. A company can find no trace of intrusion on its side while information entered on its site was read in the browser. Server logs do not see what happens in the browser.

Why the distinction matters.

  1. For visitors. They are the ones who may have been affected. Knowing whether they visited an affected page during the exposure window tells them what to watch: card statements, cryptocurrency transfers, updating their iPhone.
  2. For the law. Law 25 and PIPEDA cover unauthorised access to personal information, not only intrusion into a server. The assessment must therefore cover the browser too. We detail those obligations in the article on your duties when a third-party script steals from your customers.
  3. For trust. A statement that only covers the company's systems leaves the visitors' question unanswered. The useful answer says which pages loaded the script, during which hours, with which controls, and what visitors should check.

This applies to every organisation in the inventory. It does not establish that Equifax visitors were affected: it establishes that the question has to be asked on the browser side, and that the published statement does not settle it.

The Equifax scenario: a trusted journey

The my.equifax.ca domain deserves particular attention. Equifax Canada's pages send consumers there to sign up for myEquifax, where they get their credit report and score, and to log in. According to its FAQ, Equifax Canada collects personal information there to verify the person's identity and create the account. A consumer can arrive from Equifax's site or from another service that sends them there. It is a trusted journey: people go there to view their credit.

The hypothesis. If a page on that journey loaded the compromised Optimizely project during the exposure window, the consumer's browser received the malicious code. The risk would then depend on the page, the authentication method, the data reachable by JavaScript and the functions actually triggered.

The technical limits. A script running on an Equifax domain cannot, by default, read cookies from another domain, such as that of the service the consumer came from: the browser's same-origin policy prevents it. It could, however, read what the Equifax page shows or what is typed into it, some URL parameters, tokens exposed to JavaScript or unprotected cookies on the current domain, depending on the architecture. Arriving at my.equifax.ca from another service therefore shows no leak of that service's session.

What is not demonstrated. We do not know which pages on my.equifax.ca loaded the project, or whether track.js ran there in an authenticated session. No public evidence we reviewed supports a conclusion that information was collected or exfiltrated on that journey. Equifax says it found no evidence its own systems were compromised, which, as explained above, does not settle the browser question.

The questions to ask. Which Optimizely project IDs were tied to my.equifax.ca, and on which pages? Were those pages viewed in an authenticated session or during sign-up? Did a CSP or client-side monitoring block track.js from loading? Do the logs show requests to 1688360.cc? Have the services that send their customers to this domain asked Equifax? These are the questions a serious impact assessment should settle.

Mobile apps and smart TVs

Crave runs on Samsung and LG TVs, Android TV, Roku, Fire TV and Apple TV, according to its device compatibility page. According to BCE's results, Crave had 5.07 million subscriptions in the second quarter of 2026, with no breakdown by device. That number shows the scale of the service, not any exposure tied to the incident. The Samsung and LG platforms run web apps built in HTML, CSS and JavaScript, and a webOS app can even be hosted on a remote server.

The question is fair, the answer is not public. If a Crave app loaded a web component carrying the compromised project, the exposure would reach beyond the browsers the sweeps measured. But the architecture of the Crave apps is not public, and no source says they loaded the snippet. Precision matters here too: a script running in an app is not a compromise of the TV's system, and the payload's iOS branch targeted iPhones and iPads, not TVs. No compromise of TVs running Crave has been demonstrated. It is an attack-surface question that only Crave's operator can answer.

The limits of public visibility

Six steps are often blurred in coverage: a vendor vulnerability, the delivery of a modified script, the payload's execution, access to personal information, its exfiltration and a duty to report. Public sources establish the first 2 for the snippets found. The rest can only be established by the organisations, with their logs, their CSP violation reports and timestamped copies of what their pages served. An organisation with none of these will struggle to show the extent of its exposure, either way. The Law 25 and PIPEDA obligations are in our article on your duties when a third-party script steals from your customers.

The Sentrix angle: people first, evidence over claims

At Sentrix, we help organisations govern their risks, their vendors and their compliance, and we step in when an incident occurs. This incident sums up what we stand for.

An outage is not a cyberattack, and a cyberattack does not always look like an outage. When a service goes down, the temptation is to see an attack. This one, very real, played out while the sites worked normally: the right address, an encrypted connection, no careless move by the visitor. No fraudulent link to click, no file to download. An outage stops you from using a service. An attack like this one happens while you use it.

People at the centre of the investigation. Cybersecurity does not only protect a company's systems. It also protects the people who trust it. An investigation that stops at the servers answers the company's question, not its customers': how many visitors loaded the code, what data was reachable during their visit, what checks were done, and were the people concerned told? Those are the questions we suggest asking first.

Monitoring your credit does not protect where you use it. Credit monitoring and dark web monitoring detect exposure after the fact. Our own dark web monitoring service aims to learn of a leak before the attacker does, not to protect the page where information is entered. That protection happens upstream, in the scripts an organisation allows on its pages.

No panic, no accusation, no downplaying. The first of our principles is "evidence over claims". We do not write that an organisation lost data without proof. Nor do we write that there is nothing to see because its servers are intact.

What we do differently

  • Rank a script vendor as critical. Our third-party risk module lets you rank vendors by criticality and type of data access, and critical vendors get an in-depth review. We recommend ranking as critical any tool that runs code on a login or payment page.
  • Track a vendor incident once. The platform monitors vendor security news and breach disclosures. In our continuous threat exposure management, the supply chain is one of 6 domains, and a vendor incident is tracked there once, alongside the third-party risk register.
  • Tie compliance to real risk. One incident can touch PCI DSS, Law 25, PIPEDA and ISO 27001. In Sentrix, exposure, compliance and third-party risk share one record, and a closed exposure is logged where the auditor reads it.
  • Find out what was accessed. Our incident response service is answered 24/7 by an analyst, and its forensic investigation determines how the incident happened and what was accessed. That is exactly the question this incident leaves open.
  • Hold ourselves to the same bar. Sentrix's ISMS is certified to ISO/IEC 27001:2022, a standard whose Annex A covers security in supplier relationships.

Our recommendations

The security of a digital service no longer depends only on the infrastructure that hosts it. It also depends on the outside components it trusts, the privileges it delegates to them and its ability to see quickly when a legitimate dependency starts to misbehave.

  1. Inventory third-party scripts as critical vendors. For each application: which scripts, which owner, which purpose, on which pages. Login, payment and financial service pages first. That is the work of our third-party risk programme.
  2. Limit privileges. Fewer third-party scripts on sensitive pages, and an enforced Content Security Policy, without 'unsafe-eval', with a short list of data destinations.
  3. Monitor integrity and keep history. Detect changes to a third-party script and keep its versions, then cross-check them with logs, to rebuild what it contained on 16 November, not only today.
  4. Prepare the response. Keep the logs that let you rebuild script loads, violations and exposure windows, and be able to switch a dependency off in minutes. Our incident response service helps prepare for it.
  5. Assess the vendor on its technical privileges, not only its contract. Who can write to its code, how its accounts are protected, which surfaces it runs on. Our questions are in the article on what to ask a script vendor, and the payment obligations in the one on PCI DSS 6.4.3 and 11.6.1.

This incident does not show that every organisation on the list lost data. It shows that a technical trust relationship can be used to deliver malicious code at scale, and that without proper history it can be hard to rebuild afterwards what pages ran.

The next step

If your organisation uses Optimizely Web, open the Project JavaScript of every project and the code of every experiment today. Look for 1688360.cc, atob( and any code you did not write, then review the users, tokens and integrations that can write to it. If you want to run this exercise with us, contact us.

Sources

Frequently asked questions

Q01
Which Canada-linked organisations are in the inventory?
The inventory published by Report URI names Equifax Canada (my.equifax.ca and www.equifax.ca), Crave, CTV News, BNN Bloomberg and CP24, TSN, Hootsuite, Intelex and Starbucks Canada. The names come from a sweep of Optimizely snippets, not from notices by the companies. Being in the inventory means the snippet tied to the domain carried the code, not that every page or every visitor was affected.
Q02
Did the 178 organisations suffer a data breach?
Nothing supports that claim. A vendor vulnerability, the delivery of a modified script, its execution in a browser, access to information, its exfiltration and a duty to report are 6 separate steps. Public sources establish the first 2 for the snippets found. For the rest, they document capabilities and cryptocurrency theft, without counting the cards, cookies or devices affected.
Q03
Were Equifax Canada's customers exposed?
No public evidence we reviewed supports that conclusion. my.equifax.ca, which is in the inventory, is where Equifax Canada's site sends consumers to sign up for and log in to myEquifax. Knowing which pages loaded the compromised project, in which sessions and behind which controls, needs a technical check only Equifax can make. Equifax says it found no evidence its own systems were compromised, which says nothing about what ran in visitors' browsers.
Q04
Were the Crave apps on smart TVs affected?
No published source says so, either way. Report URI analysed websites. Crave runs on Samsung and LG TVs, whose platforms run web apps, but the architecture of the Crave apps is not public. Only Crave's operator can say whether they loaded the snippet. Even then, a script running in an app does not mean the TV itself was compromised.

Let's talk about your compliance program.

Last updated: