Dev.to Security πŸ” Cybersecurity πŸ‘ 0 πŸ“– 10 min read

178 organizations served malware through one trusted Optimizely snippet: 6 third-party script checks to put in code review in 2026

On November 16, 2026, an attacker wrote a malicious loader into the Project JavaScript field of Optimizely Web customer projects. The snippets kept arriving from a host those sites already trusted. In Scott Helme's Repo

178 organizations served malware through one trusted Optimizely snippet: 6 third-party script checks to put in code review in 2026

On November 16, 2026, an attacker wrote a malicious loader into the Project JavaScript field of Optimizely Web customer projects. The snippets kept arriving from a host those sites already trusted.

In Scott Helme's Report URI investigation, published November 16 and updated at 09:17 UTC on November 17, the team had fetched 1,611 distinct Optimizely snippets and found the loader in 188 of them, belonging to 178 organizations.

A check of that post on November 21 still showed those same figures, with no newer update.

Report URI says the real number is likely higher, because the team could only check snippets found in time. The published count may still rise. Optimizely has not published its own count of affected organizations, but its November 18 status update says fewer than 5% of its customers were impacted.

Report URI sells script monitoring and Content Security Policy reporting. The post is the primary technical account, so the figures and mechanics below come from that investigation, with official and independent coverage placed beside it.

Named organizations in the investigation's opening list include Zoom, the BBC, CNN, GitLab, and HSBC. Helme's post has the full list.

BleepingComputer independently confirmed, from urlscan.io scans and copies of the modified files, that the malicious script loaded or attempted to load on multiple named sites, including Zoom, GitLab, BBC, and CNN.

Visitors of the affected organizations received malware: a crypto address swapper, a card skimmer, cookie and screenshot theft, and iOS exploit-kit loaders. First-party application code did not have to change for any of that to run.

The six checks below are a review structure drawn from this incident, not a list published by Optimizely or Report URI. The durable habit is to treat Project JavaScript, tag configuration, and Content Security Policy as code, and to review the pull request that adds or loosens them.

What was served, and through which trusted script

Optimizely Web works by having each customer load a JavaScript snippet, usually from cdn.optimizely.com. The file contains Optimizely's code, the customer's experiments, and an optional Project JavaScript field: custom dashboard code that runs on every page.

The attacker wrote a loader into that field and left the rest of the snippet untouched. The file still arrived from Optimizely's CDN, with Optimizely's certificate, or from a customer-hosted copy of the same project.

Report URI is explicit on the delivery path. Optimizely's CDN served what it was given, and the incident was not CDN cache poisoning. Optimizely's own www and login sites were affected too.

A trusted Optimizely snippet can carry Project JavaScript that loads an unexpected host, unless a tight Content Security Policy blocks it.

Three loader shapes showed up in the infected projects. Most used a plain script tag pointing at https://1688360.cc/track.js. A second variant hid the same URL in base64 and decoded it with atob().

A third variant carried a full inline copy of the payload and ran through eval if the attacker domain was blocked. Smartsheet's snippet was cleaned and then re-infected with the base64 variant, so write access still existed after a first cleanup.

Official statements, and what is still unknown

Optimizely's status page, at 01:02 UTC on November 17, 2026, listed a "Web Experimentation Security Incident." An unauthorized party "exploited a vulnerability in Experimentation and used it to add malicious JavaScript to some customer projects."

The same note said Optimizely had fixed the vulnerability, removed the injected code, secured affected user accounts, and added protections against unauthorized code delivery.

In a November 18 update to the same item, Optimizely said the attacker exploited a flaw in the Web Experimentation account signup flow to impersonate users on targeted customer accounts. The company said the incident was not a credentials breach and was limited to Web Experimentation.

In a statement to BleepingComputer, Optimizely was more specific. The company said the vulnerability was in "the Experimentation account signup flow," and that it had blocked the vulnerable path, deployed a fix, ended sessions, revoked tokens, reset access, blocked code delivery in Cloudflare, and notified affected customers.

At the time Report URI first published, the researchers said they did not know how the attacker got write access to so many unrelated projects. Optimizely's November 18 account of user impersonation is the fullest public explanation so far, and a detailed technical postmortem has not been published.

A separate "Partial Outage Affecting Recommendations Tracking and Reporting" was also on the status page on November 17. The Recommendations item is an infrastructure incident, not the security incident.

What the payload did, and which counts are known

Report URI describes a shared track.js payload. The script could swap cryptocurrency addresses, skim card fields typed into a site's own forms, send cookies and screenshots, and load the Coruna and DarkSword iOS exploit kits.

Cards entered in hosted payment iframes, such as Stripe or Adyen, were out of reach. Cards typed into a site's own form fields were not. The investigation shows the kits were delivered to iPhones and iPads, but it does not publish a count of devices that were successfully exploited.

The payload was rebuilt 26 times during November 16. Report URI's November 17 09:17 UTC update recorded dated on-chain receipts of about $5,489 in BTC, $760 in LTC, $719 in TRON, $101 in ETH, and $82 on BNB Smart Chain.

Attackers emptied those wallets early on November 17. Some wallets had history before this campaign, as BleepingComputer also noted. Report URI is clear: there is no way to see how many cards, cookies, screenshots, or iOS devices were affected, so the on-chain total is not a complete loss figure.

By Report URI's count, 169 snippets were still infected at 20:56 UTC on November 16 and only one of the 188 (TalkTalk) at 21:52 UTC. The November 17 update adds that all observed files were clean by 23:11 UTC.

Kit comments in the investigation say iOS 18.7.3 and later, or 26.3 and later, are patched. Site owners who still use Optimizely Web should follow Helme's immediate steps: open Project JavaScript and experiment custom code, review write access, and rotate credentials.

6 third-party script checks to put in code review

Six third-party script checks for code review, with Subresource Integrity marked as a weak fit for a changing snippet.

Each check applies to Project JavaScript, experiment custom code, account access, or Content Security Policy, and to the pull request that adds or loosens a third-party script or a CSP.

  1. Treat Project JavaScript and tag-manager custom code as production code
  2. Do not take a trusted host as a standing integrity guarantee
  3. Review Content Security Policy as code, including the holes this payload used
  4. Do not rely on Subresource Integrity for a snippet that changes
  5. Look for loader tricks, not only a domain string
  6. Check the PR that adds or loosens a third-party script, CSP, or tag config

1. Treat Project JavaScript and tag-manager custom code as production code

The loader lived in a dashboard field that runs on every page, not in the customer's application repository. A team that can ship JavaScript from an experimentation or tag tool without a review has already made that path part of the production attack surface.

The review question is who can edit Project JavaScript, experiment custom code, or equivalent tag configuration, and whether that change gets the same review as application JavaScript.

2. Do not take a trusted host as a standing integrity guarantee

The malicious code arrived from cdn.optimizely.com, or from a first-party host that was only proxying the same project. A Content Security Policy that only allows the vendor host would not have distinguished the loader from legitimate snippet code.

Report URI's lesson is direct: "trusted JavaScript doesn't always remain trustworthy." The review question is how the team would notice if this trusted script changed, and what else that host can now run.

3. Review Content Security Policy as code, including the holes this payload used

Report URI crawled 256 of the affected organizations' own hosts and evaluated their policies against what the payload needed. In the published results, 68 had no Content Security Policy. 50 had a policy that still allowed the payload and the data theft.

17 hosts blocked data exfiltration with connect-src but still allowed address swapping, which happens inside the page. 24 (15%) blocked the payload, generally with a tight script-src. Treat the 256-host crawl as Report URI's sample, not as a census of the web.

Allowing cdn.optimizely.com was never going to help, because the malicious code arrived from the trusted host itself. Two settings let the attack through: 'strict-dynamic', which trusts scripts the snippet inserts, and 'unsafe-eval'.

Optimizely's own documentation tells customers to allow 'unsafe-eval', and that setting let the inline fallback run. Report URI's HSBC example shows the pattern: the policy blocked track.js and the data theft, but the inline copy still ran.

The review question is whether a pull request adds a host, 'unsafe-eval', 'unsafe-inline', or 'strict-dynamic' that a loader could use. A restrictive script-src and connect-src, plus reporting, are the controls that mattered here.

4. Do not rely on Subresource Integrity for a snippet that changes

Subresource Integrity (SRI) pins a script to a fixed hash, so the browser refuses a file that has changed. Report URI's post does not test SRI, and "CSP Integrity" in that post is the name of a Report URI product, not an SRI result.

Optimizely snippets change when experiments or Project JavaScript change, so a static integrity hash on the snippet URL is a weak fit and would break or be skipped. Subresource Integrity is not the recommended fix for this class of snippet.

The hash pin can still matter for other, stable third-party scripts in the same review. For a file that is meant to change, a tight script-src and connect-src, no custom JavaScript, or tighter vendor access is the stronger control here.

5. Look for loader tricks, not only a domain string

A search for 1688360.cc would have missed the base64 atob() variant. The inline fallback would have run even if the attacker domain was blocked, unless eval was also blocked.

The review of custom JavaScript has to look for new script tags, encoded URLs, and eval, not only for known bad domains after an incident. Access can persist after a first cleanup, as the Smartsheet re-infection shows.

6. Check the PR that adds or loosens a third-party script, CSP, or tag config

A pull request that adds a new vendor script, widens script-src or connect-src, or connects a tag manager is a security review, even when the ticket is labeled as marketing or experimentation.

First-party application code was not the breach path. The durable habit is still to review the configuration and Content Security Policy that let a trusted script run. Secure code review already treats those policy files as part of what the review covers.

What review cannot settle by itself

A pull-request checklist will not disclose a vendor vulnerability in a signup flow. Such a list will not reconstruct how many cards or devices were affected, and will not replace Optimizely's own access controls or a later postmortem.

The point is narrower: third-party script policy is code, and it should be reviewed as code before it runs on every page.

Review third-party script policy the way you review application code

Qodo, the AI code review platform, reviews pull requests with full codebase context, and a team's custom instructions can ask it to flag changes that add or loosen third-party scripts or Content Security Policy.

In the Qodo portal, under Configurations, Custom instructions, or in .pr_agent.toml under [review_agent], a team can add instructions for new script hosts, 'unsafe-eval', and 'strict-dynamic'. Useful keys under [review_agent] are issues_user_guidelines and compliance_user_guidelines.

The same checks can live in a pr_compliance_checklist.yaml file at the repository root. Teams can also apply the same custom instructions once, at workspace or Git organization level. Qodo flags; your team decides.

Dashboard-only Project JavaScript still has to be captured in versioned configuration before a pull-request review can see it. The review habit covers the change that lands in the repository, including Content Security Policy and script-include files.

Qodo is the AI code review platform with high-precision PR review, full codebase context, enforceable rules, and local review inside your coding agents.

FAQ

How many organizations were affected?

In Scott Helme's Report URI investigation, published November 16 and updated at 09:17 UTC on November 17, the team had found the loader in 188 of the 1,611 snippets it checked, belonging to 178 organizations. Report URI says the real number is likely higher, so the published count may still rise.

Was this a CDN cache-poisoning attack?

No. Report URI states that Optimizely's CDN served exactly what it was given.

What did Optimizely say?

The status page describes a vulnerability in Experimentation, now fixed, with injected code removed. Optimizely's statement to BleepingComputer and its November 18 status update both name the account signup flow.

Do we know how the attacker got access?

Optimizely says the attacker exploited a flaw in the Web Experimentation account signup flow to impersonate users on targeted customer accounts. A detailed technical postmortem has not been published.

Did Content Security Policy help?

In Report URI's 256-host crawl, 24 (15%) blocked the payload. 68 had no Content Security Policy.

Can Subresource Integrity pin an Optimizely snippet?

Subresource Integrity is a weak fit for a file that is meant to change. Report URI did not test it, and it is not the recommended fix for this class of snippet.

Should Report URI be treated as an independent source?

Report URI published the primary investigation, and the company sells script monitoring. Official wording comes from Optimizely's status page. Independent confirmation of loads on named sites comes from BleepingComputer.

πŸ“° Read the original article on Dev.to Security

Originally published by Dev.to Security. Aggregated on AIWithGhost for educational purposes β€” full credit and traffic to the original publisher.