At 04:21 UTC on Monday, a BBC page began loading a script from a domain called webshell.lol. An hour later, 1688360.cc was registered. By 08:28, code from that new domain was running on bbc.com, served by Optimizely’s own content delivery network, signed with Optimizely’s own certificate, through the same script tag the BBC had used for years.

By the end of the day, Scott Helme and the team at Report URI had found the injected code in 188 of the 1,611 Optimizely snippets they checked. The snippets belonged to 178 organizations, including Zoom, CNN, Fox News, Salesforce, GitLab, HSBC, IKEA, Equifax, Air France-KLM, Ledger and UNICEF USA.

Banking and finance was the largest affected sector, with 32 organizations. Technology and SaaS accounted for 23, retail and fashion 20, streaming and entertainment 12, news and media 10. None of those companies had been breached directly. They had all outsourced a piece of their page to a vendor, and the vendor’s platform was used to hand their visitors a malware kit.

What changed: Optimizely’s security advisory, updated on 19 November, says the attackers used a flaw in the Web Experimentation signup flow to impersonate users on targeted customer accounts, that fewer than 5% of customers were affected, and that the company detected, blocked delivery and closed the vulnerability within three hours of detection. It says the incident was not a credentials breach and that no malicious code remains. The company still has not explained how a signup-flow flaw allowed an outsider to write code into other customers’ projects. Salesforce, one of the affected customers, says it has disabled Optimizely scripts indefinitely.

What Optimizely actually does

Optimizely Web Experimentation runs A/B tests. Each customer adds one script tag to its site, which pulls a per-project snippet from cdn.optimizely.com. That snippet holds Optimizely’s code, the customer’s experiment definitions, and an optional field called Project JavaScript: custom code a customer writes in its own dashboard, which runs on every page load.

The attacker wrote a loader into the Project JavaScript of customer projects. Nothing else in the snippets changed. The file still came from the host every affected site already trusted, which is why no allowlist keyed to Optimizely’s domain would have helped. Report URI was explicit that this was not cache poisoning. The CDN served exactly what it was given.

The loader came in three forms. Most snippets used a plain script tag pointing at https://1688360.cc/track.js. Some hid the same URL in base64 and decoded it with atob(), so a text search for the domain would miss it. A third version embedded the full payload and ran it through eval() if the remote script was blocked. Smartsheet’s snippet was cleaned at one point and then reinfected with the encoded variant, which suggests the attacker still had write access hours into the cleanup.

Some customers serve the snippet from their own subdomain, such as opt.scheels.com. Those were affected too, because the content originated in the same Optimizely project. So were optimizely.com and login.optimizely.com.

One payload, four jobs

Every visitor got the same track.js, and every capability in it ran at once. What happened next depended on the device and on what the visitor did.

The crypto clipper was live from the start. It scanned page text, form fields, links, images, the clipboard and outgoing requests for valid wallet addresses across Bitcoin, Ethereum and EVM chains, TRON, Solana, TON, Litecoin, Bitcoin Cash and Dogecoin. It validated checksums so it only swapped real addresses, then substituted the attacker’s. It re-scanned the page as it changed. A visitor who copied an address from an affected page, or from anywhere else while the tab stayed open, could paste a different address than the one they copied.

Card skimming was added at 13:43 UTC. The script watched form submissions and every fetch or XHR POST for a card number that passed a Luhn check, an expiry date and a CVV. With all three, it sent them to 1688360.cc/card.php using navigator.sendBeacon. Cards entered into hosted payment iframes from Stripe, Adyen and similar providers were out of reach. Cards typed into a site’s own form fields were not.

Cookie and screenshot theft ran alongside. When the script swapped an address near wallet or payment wording, it also sent the page URL, any cookies not marked HttpOnly, the addresses it had replaced, and a screenshot rendered with html2canvas.

On iPhones and iPads, the script loaded boot.js from two more attacker domains, read the iOS version, and opened an exploit chain in a hidden iframe. Devices on iOS 13 through 17 received Coruna. Devices on iOS 18 received DarkSword. A third chain for iOS 26 only fingerprinted the device. The Coruna loader was byte-identical to a sample Report URI had analysed in October from a different campaign. Pages on iOS 18.6 or earlier reloaded every 60 seconds to retry, which would have been a visible symptom.

The kits are the most serious part of the incident, and the least confirmed. Public scans show the exploit pages being delivered. Nobody has published evidence that a device was actually compromised. The kits’ own comments say iOS 18.7.3 and later, and 26.3 and later, are patched.

The money, and the timeline

By 19:25 UTC, the attacker’s wallets had received about $7,040 in roughly nine hours. Bitcoin accounted for $5,380 of it, with smaller amounts in Litecoin, TRON, Ethereum and BNB Smart Chain. Funds were swept to exchanges between 13:03 and 13:05. The TRON wallet had been in use since March 2026, so this actor was active long before this campaign.

Those figures are on-chain receipts only. There is no way to know how many cards were skimmed, how many cookies and screenshots left the affected pages, or how many devices were exposed.

Report URI recorded 26 rebuilds of the payload during the day. Features and wallet addresses were added; none were removed. The first injection appeared at 04:21, mass cleanup ran from about 20:40 to 21:10, and every file the team tracked was clean by 23:11. For most sites that means roughly 16 hours of serving malware, and Optimizely’s own write-up puts the worst case near 19 hours.

The campaign did not begin on Monday. Between 13 and 16 November, the same actor injected webshell.lol/get.js through Kissmetrics, StorifyMe and Minute Media scripts on sites including securitybank.com, faz.net, smh.com.au, vidaxl and foxsports.com. StorifyMe has since confirmed that unauthorized JavaScript reached some of its customers’ integrations. Optimizely was the escalation, not the opening move.

The infrastructure came down quickly once it was public. 1688360.cc was put on clientHold at 00:37 UTC on Tuesday and stopped resolving. webshell.lol followed at 03:37, one of the two iOS hosts stopped resolving, and the other stopped responding. That protects future visitors. It does nothing for anyone who pasted an address on Monday.

What Optimizely now says, and what it still does not

Optimizely’s account has become more specific since the first day. In its 19 November advisory, the company says an unauthorized third party “exploited a flaw in the Optimizely Web Experimentation account signup flow” and “used it to impersonate users on targeted customer accounts and insert malicious JavaScript into those customers’ projects.” It says it detected the activity, blocked delivery, closed the vulnerability within three hours of detection, removed the injected code, ended active sessions, revoked tokens and reset access for affected accounts. It also says the incident was not a credentials breach and was limited to Web Experimentation.

That answers one question and leaves a larger one open. The company has now confirmed that the entry point was the signup flow and that the effect was impersonation. It has not explained how a flaw in signup allowed an outsider to impersonate existing users, whether email verification, single sign-on or session handling was involved, or how access to one account translated into write access to the projects of 178 organizations. “Targeted customer accounts” suggests the attacker chose victims, but the company has not described the mechanism.

The “not a credentials breach” line is technically important and practically narrow. It means the attacker did not steal customer passwords. It does not mean the attacker had legitimate access. Impersonation is still account takeover in effect, even if the credential was never exposed.

Forrester’s analysts reached a similar reading: the attacker impersonated administrator users and modified Project JavaScript without stealing customer passwords or compromising an Optimizely administrator account. That is a supply-chain failure inside the vendor’s own onboarding path, not a phishing campaign against 178 companies.

Three hours, or sixteen?

Optimizely’s containment claim and Report URI’s timeline measure different things, and the difference matters for anyone trying to scope exposure.

Optimizely says it closed the vulnerability within three hours of detection. It does not say when detection occurred. Report URI observed the first malicious injection at 04:21 UTC and the last tracked file clean at 23:11 UTC. Optimizely’s first public status update came at 01:02 UTC on 17 November, and its advisory was updated on 19 November. The company has not published a minute-by-minute timeline that reconciles the three-hour containment window with a delivery window that lasted most of the day.

Both things can be true: the vulnerability may have been closed within three hours of someone noticing, while already-injected code continued to be served from caches and customer projects until it was removed. The point is that “contained in three hours” is not the same claim as “visitors were exposed for three hours.” Anyone estimating their own exposure should use the delivery window, not the containment claim.

The scope numbers have the same shape. “Less than 5% of customers” is a share of Optimizely’s customer base. Report URI’s 178 organizations is a count from 1,611 snippets the team could find and test, not a census of Optimizely’s customers. The two figures are not necessarily in conflict, but they cannot be checked against each other without the denominator Optimizely has not published.

What customers are doing

Salesforce, named in Report URI’s list, said on 18 November that it had disabled Optimizely scripts across preview and production instances. It said it had no evidence that the incident affected Salesforce-platform or subscription services, and that the scripts would stay off until Optimizely’s remediation was verified, with no timeline to re-enable. That is a notable choice: even after the vendor said the flaw was closed, a large customer decided the safe state was off.

Other customers have paused the platform while they investigate. TPG Telecom, which owns Vodafone Australia, said it found no evidence that Vodafone systems or customer data were compromised and that the affected pages had no payment or login functions. HSBC said unauthorized scripts were not active on its websites; Report URI says the bank’s Content Security Policy blocked the data theft, although a backup copy of the code still ran.

The common thread is that customers are not treating the vendor’s all-clear as the end of the incident. They are checking their own pages, their own policies and their own logs.

Why almost nobody was protected

The most useful part of Report URI’s work is not the list of victims. It is the measurement of who blocked the attack and why.

Helme’s team took 256 affected organizations’ own hosts, evaluated their Content Security Policy against what the payload needed, and got usable policies for 159 of them. Sixty-eight had no CSP at all. Fifty had a policy that allowed both the payload and the data theft. Seventeen let the payload run but blocked the outbound theft through connect-src; address swapping still worked on those sites, because it happens entirely inside the page. Twenty-four, about 15%, blocked the payload outright.

Allowing cdn.optimizely.com was never the fix, since the malicious code arrived from that host. What separated the protected sites was a restrictive script-src that listed only the hosts they needed, so 1688360.cc was never permitted to load.

Two common settings undid otherwise reasonable policies. 'strict-dynamic' means any script a trusted script creates inherits its trust, so the Optimizely snippet handed trust straight to the attacker’s domain and host allowlists were ignored. 'unsafe-eval' let the inline fallback run even where track.js was blocked. Optimizely’s own CSP guidance recommends 'strict-dynamic' and its full sample policy includes 'unsafe-eval', because some features need it. One large bank blocked both the remote script and the data theft and still ran the inline copy.

The pattern is familiar from other third-party compromises. Earlier this year, attackers reached nearly 50 organizations through a third-party extension for ConnectWise’s ScreenConnect remote management tool, as we reported in our coverage of the ScreenConnect extension attacks. A separate campaign then used ScreenConnect’s own Run command feature against managed service providers, which we examined in our report on the SCKit malware. In each case the victims’ own code was fine. The trust they placed in someone else’s was the vulnerability. Our Security hub collects the rest of our coverage on third-party and supply-chain incidents.

What would change the picture

Four developments would materially change this story. A forensic report that traces the signup-flow flaw and the authorization path would answer the central question. A CVE or a regulatory finding would give the flaw a fixed identity and a severity. Evidence that a device was actually compromised by Coruna or DarkSword would turn a delivery attempt into a confirmed exploit. And a reinfection of a cleaned project would show that the attacker’s access survived the remediation.

As of 21 November, none of those had been published. Optimizely says no malicious code remains and that changes to the JavaScript are no longer possible. Helme said he would wait to see whether cleaned sites were reinfected before treating the flaw as closed. We found no public report of reinfection, which is not the same as proof that none occurred.

What customers should do

Optimizely says the injected code has been removed, so the remaining work is verification, damage control and configuration. Helme’s list is specific. Open every project’s Project JavaScript and every experiment’s custom code and look for 1688360.cc, atob(, and anything nobody on the team wrote. Review who and what has write access to each project, including users, API tokens and integrations, and rotate credentials.

Any card entered into a site-hosted payment form on 16 November should be treated as exposed. Any session token or secret stored in a cookie without HttpOnly should be rotated. Visitors who made a crypto transfer that day should check that the address they pasted matches the one they intended to pay. iPhone and iPad users should be on iOS 18.7.3 or later, or 26.3 or later.

Then there is the harder fix. A restrictive script-src, a tight connect-src, and no 'unsafe-eval' unless a feature genuinely requires it. Report-only mode and violation reporting would have flagged 1688360.cc within minutes without breaking the site, and it costs nothing to leave on. Customers should also decide whether a third-party experimentation script needs to run on payment and login pages at all.

Bottom line

This was not a sophisticated exploit chain against 178 companies. It was one flaw in a vendor’s signup flow, one custom-code field, and a delivery path that every customer had already approved. The payload was competent and the timeline was fast, but the reason it reached the BBC, Zoom and HSBC is that those sites deliberately execute code from a third party on every page load.

Optimizely’s revised account is more specific than its first statement, and it deserves credit for confirming the entry point and the impersonation. It still leaves the most important question unanswered: how did a signup flaw become write access to other customers’ projects? Until that is explained, the three-hour containment claim and the five-percent scope claim are assertions to check, not facts to rely on.

That arrangement is not going away. Tag managers, analytics, A/B testing and chat widgets all work the same way. What this incident shows is that the only control that consistently held was the one the customer owns: a policy that limits where code can come from and where data can go. Fifteen percent of the sites Report URI checked had that. The rest served malware with their own domain in the address bar.

What we added: We updated this article on 21 November with Optimizely’s 19 November advisory, separated the three-hour containment claim from the roughly 16-hour delivery window, examined the “less than 5%” scope figure against Report URI’s 178-organization count, added Salesforce’s and StorifyMe’s responses, and added the developments that would change our assessment. The original technical analysis of the payload and the CSP measurements is unchanged.

Last reviewed: 21 November 2026.

Sources: Report URI, “World’s largest brands infected with malware via Optimizely” (16 November 2026); Optimizely security advisory (updated 19 November 2026); Optimizely status page (updated 18 November 2026); Salesforce, “Our response to the Optimizely ‘OPTIMJACK’ supply chain compromise” (18 November 2026); Forrester analysis (20 November 2026); CMSWire (18 November 2026); Cyber News Centre on Australian customers (20 November 2026); BleepingComputer (16 November 2026); The Record (16 November 2026); StepSecurity (16 November 2026); Optimizely World, “Lessons From the Optimizely Web Experimentation Compromise” (17 November 2026).

Have a tip? Email [email protected] or use the contact form and choose Editorial. We do not promise anonymity.