BlogIncident Analysis

The Optimizely Supply-Chain Attack: How 178 Brands Served Malicious JavaScript to Their Own Visitors

On 16 November 2026, attackers exploited a vulnerability in Optimizely's Experimentation account signup flow to inject JavaScript into 188 customer snippets served through Optimizely's trusted CDN. Within hours, visitors to 178 organizations — including BBC, CNN, Fox News, Zoom, Salesforce, GitLab, IKEA, HSBC, KLM, Air France, American Eagle, UNICEF USA, and many more — were being served a crypto-wallet clipper, a payment-card skimmer, and an iOS exploit-kit loader. None of those organizations were themselves compromised. They were serving the attacker's code because the vendor they trusted for A/B testing was.

11 min readBy TigerGate Research, Security Research TeamPublished Updated

TL;DR

  • · One vulnerability in Optimizely's account signup flow let attackers write JavaScript into other customers' Experimentation Projects without owning those accounts.
  • · Injected code was served through Optimizely's own CDN, so it was trust-chained into every visitor of every affected site. Roughly 188 infected snippets / 178 organizations.
  • · Three payloads: a crypto wallet clipper (ETH / Tron / BTC), a card skimmer, and an iOS-only loader for the Coruna and DarkSword exploit kits.
  • · Short window — infected snippets first captured at 08:08 UTC, all cleaned by 23:11 UTC the same day — but the trust model takes longer to recover.
  • · If you embed any third-party script on a page that handles customer data or payments, you need the controls described below: SRI, CSP, allow-listed first-party vendors, change-detection on third-party JS.

What happened

Optimizely is one of the largest A/B-testing and experimentation platforms on the web. The product works by giving customers a small bootstrap JavaScript snippet to paste into their site's <head>; that snippet, served from Optimizely's CDN, fetches each customer's Project configuration and executes any "Project JavaScript" the customer has defined — code that runs on the live production site for all visitors matching an experiment.

On the morning of 16 November 2026, security researcher Scott Helme of Report URI began noticing unfamiliar code appearing inside Optimizely-served scripts across his monitored targets. Scanning 1,611 snippets he had baseline coverage of, he identified 188 infected files tied to 178 distinct organizations. Optimizely confirmed within hours that the attacker had exploited a vulnerability in the Experimentation account signup flow to plant the JavaScript — with no credentials for, and no access to, the victim customers' Optimizely accounts.

By 23:11 UTC the same day Optimizely had closed the vulnerable path, revoked tokens, ended active sessions, blocked delivery at Cloudflare, and confirmed all 188 snippets were cleaned. On the blockchain side, the attacker's wallets received roughly $6,600 – $7,040 in a mix of Bitcoin, TRON/USDT, and Litecoin before the campaign ended — not a huge haul for the scale, which tells its own story about what this was really an exercise in.

How the signup-flow bug became JS on 178 brands

Most supply-chain stories end at "the vendor got breached." This one is interesting because Optimizely's production customer accounts were not compromised. The attacker did not phish an Optimizely employee, did not reset any customer's password, and did not log in as any customer. They abused the account-creation path.

The precise bug has not been fully published, but the shape is clear from Optimizely's own statement and the observed exploitation pattern: the signup flow allowed the new-account request to associate with, or write into, resources that should have been scoped to existing accounts. In practice that gave the attacker a write primitive against the Project JavaScript field of other customers' projects. Once written, that JavaScript was served by Optimizely's CDN to every visitor whose session landed in the affected experiment — because there is no distinction, at the CDN layer, between "legitimate JS the customer wrote" and "JS someone else wrote into the customer's project."

The structural lesson: a privilege-boundary bug anywhere in a vendor's control plane is equivalent to a compromise of every one of that vendor's customers' sites. The attacker does not have to be inside the customers to be inside the customers' pages.

The three payloads

1. Crypto wallet clipper

The main-line payload. The injected code scans the rendered page for anything that looks like a cryptocurrency address — Ethereum, TRON, and Bitcoin formats specifically — and replaces it with a hard-coded attacker-controlled address. It achieves this through two complementary mechanisms:

  • Clipboard API interception. The script wraps navigator.clipboard.writeText and the older document.execCommand('copy') handlers so any address a user copies off the page is silently swapped to the attacker's address before it lands on the clipboard.
  • DOM rewriting via MutationObserver. As addresses appear in rendered text (e.g. a checkout receipt, a donation page, a crypto-exchange UI, a wallet balance display), the observer rewrites the displayed value to the attacker address — so even users who type rather than paste still see the replaced value.

Hard-coded attacker wallets received roughly 0.08298 BTC (~$5,380), $719 in TRON/USDT, and $762 in Litecoin by end-of-day 16 November — total ~$7,000.

2. Payment-card skimmer

Documented in later variants of the payload. Standard Magecart-shape JS: attach event listeners to form fields matching patterns for card number, expiry, and CVV; capture on blur or submit; exfiltrate via a POST to attacker infrastructure. On e-commerce and donation flows operated by the 178 brands, this was the real revenue line — much larger than the crypto-clipper take if it ran for a full day.

3. iOS exploit-kit loader

A targeted branch of the payload, gated on navigator.userAgent matching iOS. On a hit, the script dynamically imports boot.js from 5vobauua3ojpklx[.]icu, which selects between the Coruna and DarkSword commercial iOS exploit kits. The exploit-kit loader is the part that tips the campaign over from "opportunistic cryptocurrency theft" into "the operators also had access to high-end mobile exploitation" — which is a different-order threat-actor profile.

Infrastructure indicators

Hunt these across your egress logs, CDN logs, and browser reports (CSP report-uri, SRI violation reports) for the window of 16 November 2026 onward:

IndicatorValue
Payload host1688360[.]cc
Payload path/track.js (19,513 bytes)
iOS loader5vobauua3ojpklx[.]icu/boot.js
SHA-256 of track.js1ec4245b3b303304ea24a89b3b2351b0f0c085087050f517d23dcba9694c165a

Who was affected

Confirmed in public reporting so far — a partial list from the 178-organization total:

  • News & media: BBC, CNN, Fox News
  • Communications & collaboration: Zoom, GitLab
  • SaaS & cloud: Salesforce
  • Retail & e-commerce: IKEA, American Eagle
  • Travel: KLM, Air France
  • Financial services: HSBC, HSBC Insurance
  • Non-profit & civic: UNICEF USA, Change.org
  • Infrastructure & links: Bitly

The full 178 includes long-tail e-commerce, publishing, and marketing sites that either have not individually disclosed or have not been named in public write-ups.

Timeline

Date / UTCEvent
16 Nov, 08:08:15Internet Archive captures first malicious track.js.
16 Nov, 08:10First reported build timestamp of the payload.
16 Nov, late morningReport URI's Scott Helme identifies 188 infected snippets across 178 organizations after sweeping 1,611 known scripts.
16 Nov, 21:52Optimizely has cleaned all but 1 of 188 files; vulnerable signup path blocked; tokens revoked; Cloudflare delivery blocked.
16 Nov, 23:11All infected snippets confirmed clean.

Why this shape of attack is becoming the norm

The Optimizely incident is a textbook third-party-JavaScript supply-chain compromise, the same class as the historic Magecart-via-Inbenta and British Airways cases. The reason this shape keeps recurring is structural:

  • The web's trust boundary ends at the browser. Any script included on your page — from any vendor — runs with the same origin-level privileges as your own code. There is no per-vendor sandbox by default.
  • Vendor control planes are high-value, low-visibility targets. The customer-signup path, the admin API, the project-config endpoint — bugs in any of them can propagate to thousands of customer sites in minutes.
  • A/B testing, tag managers, analytics, chat widgets, feature flags, consent banners are a huge, permanent delivery surface. Replacing one is a six-month project; patching one requires trusting the vendor.
  • Browser-side controls (SRI, CSP, Trusted Types) are opt-in and under-deployed. The sites that would have detected injected JS fastest are those that had changed-asset reports flowing — and relatively few do for third-party experimentation scripts.

Detection and cleanup for website operators

Immediate checks

  1. Inventory your Optimizely footprint. Which properties embed the Optimizely snippet? Which Projects? When was the last legitimate change to each Project's JavaScript?
  2. Diff your current snippets against a known-good baseline. Compare the body of each Optimizely-served file against a clean snapshot captured before 16 November 2026. Any delta you did not author is in scope.
  3. Grep your CDN / egress / browser-report logs for the IOCs above — 1688360[.]cc, 5vobauua3ojpklx[.]icu, and the SHA-256 of track.js.
  4. Check your CSP report-uri and SRI-failure stream for the 16 November window. Any violation involving an Optimizely-served asset is incident-grade until proven otherwise.
  5. Visitor-impact assessment. If your Optimizely-served pages handle payments, cryptocurrency flows, or any form that collects card data, assume some fraction of your visitors were skimmed. Coordinate with your fraud, payments, and legal teams — breach-notification clocks may apply even though you did not store the exfiltrated data.

Structural fixes that would have blocked this class

  • Content Security Policy with a narrow script-src allow-list that includes only your approved first- and third-party origins. The injected loader (5vobauua3ojpklx[.]icu) would have been blocked at the browser.
  • Subresource Integrity (SRI) hashes for every third-party script that has a stable version. SRI is less useful for continuously-rotated A/B scripts, but it is the right default for pinned analytics and tag-manager snippets.
  • Trusted Types to block script-injection via innerHTML, document.write, and eval inside your own code and — through CSP policy — in third-party code.
  • A separate origin / iframe sandbox for third-party experimentation and analytics scripts, so a malicious experiment cannot read payment forms on the main origin.
  • Continuous third-party JS change-monitoring. A simple cron that fetches every third-party script you include, hashes it, and alerts on unexpected change is enough to catch supply-chain injections within minutes. Report URI is a commercial version; a homegrown version is a weekend's work.
  • Vendor security questionnaire + control-plane posture. Ask your experimentation / tag-manager / analytics vendor: do they segment customer accounts from the signup surface, do they have control-plane WAF rules, do they publish an SRI-friendly manifest, do they have a documented incident-communication SLA?

Broader supply-chain lessons from a busy fortnight

Optimizely is not an isolated incident. In the preceding two weeks we have also seen:

  • wfx-loot — 4,837 compromised GitHub accounts pushing a secret-stealer into 61,764 repositories, which exfiltrated secrets to AWS/GCP/Azure and planted long-lived cloud persistence.
  • GitBook — AI-assisted attackers breached an internal tool, exposing every customer's integration tokens and Git Sync secrets.
  • WorkOS — ARTEX-class agentic attack on a sales-side internal tool exposing CRM data and customer Slack excerpts.

The pattern is the same shape every time: your weakest link is a vendor's smallest internal tool, control-plane endpoint, or signup path — not your production code. That is the point where defender investment is going to flow in 2027.

See the SaaS supply-chain response playbook for the six-phase, 72-hour runbook to use when the next vendor bulletin lands.

How TigerGate helps

  • Attack Surface Management enumerates every third-party JavaScript origin in your pages — Optimizely, VWO, Segment, Google Tag Manager, Hotjar, Intercom, Drift, Sentry, Datadog RUM, and the long tail you forgot about — so you have a ground-truth inventory before the next vendor bulletin.
  • Website Security continuously fetches, hashes, and diffs every third-party script you include and alerts on unexpected change — the cheap control that would have caught Optimizely in minutes.
  • VAPT AI Engine scheduled sweeps against your sites validate CSP, SRI, and Trusted-Types posture per page and flag domains referenced in high-risk IOC feeds — including the 1688360[.]cc / 5vobauua3ojpklx[.]icu set from this campaign.
  • eBPF runtime detection catches outbound connections from server-side workloads to known skimmer / exploit-kit infrastructure, catching server-side variants where the attacker moves off the browser entirely.
  • Compliance evidence maps the third-party-JS control set to PCI DSS 4.0 requirements 6.4.3 and 11.6.1 (inventory and change-detection on payment-page scripts) — a brand-new 2025 requirement that this attack would have caught exactly.

Sources

  • Report URI / Scott Helme — initial research and 178-organization sweep, 16 November 2026
  • Optimizely — public statement confirming the signup-flow vulnerability, token revocation, and 23:11 UTC remediation
  • Public blockchain telemetry for the attacker wallets (BTC / TRON / LTC), 16–18 November 2026
  • Internet Archive capture of malicious track.js at 08:08:15 UTC on 16 November 2026
  • CVE tracking pending at NVD

Know every third-party script running on your pages

14-day trial. TigerGate inventories every third-party origin loaded by your public sites, continuously hashes each script, and alerts on unexpected change. So the next Optimizely-shape incident is a 5-minute alert, not a 15-hour blind spot.