An iPhone on iOS 18.6 or earlier opening an affected page on CNN, the BBC, or Zoom late on November 16 reloaded it every 60 seconds. Report URI, a service monitoring websites for unexpected script changes, found each reload gave a hidden exploit kit another attempt to break into the phone. The script making those attempts arrived through Optimizely, a New York-based platform those sites use for A/B testing, which shows different visitors different versions of a page to measure which performs better.
Report URI found malicious JavaScript in 188 Optimizely snippets belonging to 178 organizations. The list includes Zoom, the BBC, CNN, Salesforce, IKEA, HSBC, Equifax, Air France-KLM, and Ledger. The script swapped cryptocurrency wallet addresses and was built to skim payment cards and steal cookies and screenshots. On iPhones and iPads, it called the Coruna and DarkSword iOS exploit kits. Report URI's Scott Helme published the findings the same day, after a sweep ending at 19:59 Coordinated Universal Time (UTC) found 186 of the infected snippets.
The exploit code came from two iPhone kits first used by surveillance vendors and their customers and now sold together as a commercial product. On November 16, a flaw in Optimizely's sign-up flow let a threat actor push a payload to those organizations' visitors for most of a day.
Optimizely says the threat actor created accounts through that flaw and used them to impersonate existing users, guessing their email addresses by brute force. The wallets it planted took in a few thousand US dollars, a figure covering only the address swapping. Three other script vendors carried the same iPhone router, and no one has named the threat actor behind any of them. Content Security Policy would have stopped the payload at only about 15 percent of the organizations Report URI tested.
Operation Timeline
Trace the events, turning points, and connections behind the analysis.
OPTIMJACK Timeline
Key events from the first sighting of the injector on Turkish websites on September 17, 2026, to the Optimizely status update on November 18, 2026.
View Timeline Hide Timeline
-
September 17, 2026
-
October 4 to November 5, 2026
-
November 4, 2026
-
November 9, 2026
-
November 12, 2026
-
November 13, 2026
-
November 13, 2026
-
November 16, 2026
-
November 16, 2026
-
November 16, 2026
-
November 16, 2026
-
November 16, 2026
-
November 16, 2026
-
November 16, 2026
-
November 16, 2026
-
November 16, 2026
-
November 16, 2026
-
November 16, 2026
-
November 16, 2026
-
November 17, 2026
-
November 17, 2026
-
November 17, 2026
-
November 18, 2026
Sign-Up Flaw Handed Out Write Access to Customer Projects
Optimizely Web works by loading a JavaScript snippet from Optimizely's content delivery network (CDN) on a customer's pages. Each snippet carries Optimizely's own code, the customer's experiments, and an optional field called Project JavaScript, holding custom code the customer writes in the Optimizely dashboard. Code in that field runs on every page. Report URI found the threat actor wrote a loader into the Project JavaScript field of customer projects and left the rest of each snippet untouched.
The file still came from cdn.optimizely.com, served over Hypertext Transfer Protocol Secure (HTTPS) under Optimizely's own certificate, from a host every affected site already trusted. Report URI stresses the CDN served exactly what it was given, so cache poisoning played no part. Customers serving the snippet from their own domains, such as Scheels and Vodafone Australia, carried the code too, because the content comes from the same project. Optimizely's own optimizely.com and login.optimizely.com snippets were also infected.
Optimizely's incident notice confirmed the attack at 01:02 UTC on November 17, saying an unauthorized party exploited a vulnerability in Experimentation to add malicious JavaScript to some customer projects. In a fuller statement to BleepingComputer, Optimizely placed the flaw in its Experimentation account sign-up flow. It says it blocked the vulnerable path, deployed a fix, and removed the injected code. It also ended active sessions, revoked tokens, reset access for affected accounts, blocked any code from being delivered in Cloudflare, and notified all affected customers.
Both Report URI and software supply chain security company StepSecurity asked how a sign-up flaw gave an unauthorized party write access to unrelated customer projects. Optimizely's interim report to customers, dated November 17, answers in part.
Accounts the threat actor created through the sign-up flow could act as existing users on customer accounts through a combination of public information and brute force. Acting as administrator users, the threat actor then edited Project JavaScript. Optimizely says the threat actor guessed variations on customer employees' names as email addresses, up to thousands of times for each customer, and access to one customer account gave it none to another. Optimizely says the flaw bypassed authentication and no passwords were stolen or used.
Optimizely has not described the flaw itself, and its final root-cause analysis awaits an independent investigation by Mandiant. It also did not answer The Record about the specifics of Report URI's study. BleepingComputer notes it remains unclear how long the vulnerability had been present. Optimizely's report does not say either, though it dates the first impersonated user account to 00:37 UTC on November 16.
Optimizely gives no victim count. Its November 18 status update says less than 5 percent of its customers were impacted, and its report says it cannot disclose its customers. Report URI checked only the 1,611 snippets it could find in time, so the real count of affected organizations is likely higher.
The company has been breached before, as SecurityWeek notes. In February 2026 it disclosed a voice-phishing intrusion reaching its Zendesk support system, records in its Salesforce customer relationship management (CRM) system, and some internal documents. Optimizely said the threat actor could not escalate privileges, install software, or create backdoors.
Report URI's timeline puts the first Optimizely injection at 04:21 UTC on November 16, when the BBC's snippet began loading code from webshell[.]lol. The payload domain 1688360[.]cc was registered at 05:23 UTC. Mass cleanup ran from about 20:40 to 21:10 UTC, and TalkTalk's snippet was the last still infected at 21:52 UTC. Report URI saw every file clean by 23:11 UTC, and 1688360[.]cc was put on hold at 00:37 UTC on November 17.
Optimizely's timeline differs. It dates the first malicious code to 07:51 UTC. It says it detected the activity at 13:34 UTC, blocked malicious outbound code at Cloudflare at 16:06 UTC, closed the flaw at 16:10 UTC, and removed the injected code from all affected projects by 22:24 UTC. It says the 16:06 UTC block, based on known patterns, fully blocked delivery. Report URI's evening sweep still found 186 infected snippets, and no source reconciles the two accounts.
The loader of OPTIMJACK, the name StepSecurity credits to Amirul Cyber gave the campaign, came in three variants, according to Report URI. Most snippets carried a plain script tag for track.js. Smartsheet's snippet was cleaned and then reinfected with one hiding the payload's web address in Base64, so searching for 1688360 missed it.
Report URI reads the reinfection as proof whoever wrote the code still had access. The third variant embeds a lighter payload inline, and infostealer intelligence company Hudson Rock found the same 24,392-byte copy in the BBC and HSBC United Arab Emirates snippets.
The embedded copy runs locally if the remote file errors or fails to load within 2.5 seconds. Blocking the threat actor's domain therefore can leave the address swapping working. Hudson Rock advises containment by verifying the injected project code is gone and checking the bytes visitors actually receive, since a network block alone stops only the remote requests.
Four Thieves in One Script
Every visitor loading the remote file at a given moment received the same build of track.js, with every module in that build active at once. The first module is a crypto clipper holding replacement addresses for eight cryptocurrencies. Report URI says it validates checksums before swapping. Hudson Rock found checksum checks cover several address families, while Ethereum-style addresses are matched by format and Solana addresses by heuristics. Each replacement is a valid wallet address.
The clipper rewrites addresses in page text, form fields, links, the clipboard, outgoing requests, and the iOS share sheet. StepSecurity found it replaces the browser's Function.prototype.call and apply, rewriting any string resembling a wallet address before it reaches a function. An address copied from an affected page, or from anywhere else while the tab stays open, can paste as the threat actor's address. Report URI's timeline shows Ledger, a hardware wallet maker, serving the infected snippet until its cleanup at 20:41 UTC.
Report URI dates the card skimmer's addition to track.js to 13:43 UTC. It watches form submissions and outgoing requests for a Luhn-valid card number, an expiry date, and a security code, then beacons all three to 1688360[.]cc/card.php. It cannot read card fields inside a payment provider's cross-origin frame, such as those from Stripe or Adyen.
Cards typed into a site's own form fields were within its reach, and StepSecurity notes the skimmer leaves each captured card in the victim's sessionStorage. Card skimming is an established eCrime business. My analysis of one threat actor pointing artificial intelligence (AI) agents at online retailers found a campaign tied to card skimmers on at least 119 sites.
A third module fires after a wallet swap. It renders the visible page with html2canvas and posts the image to hit.php, along with the page address, up to 8,000 characters of readable cookies, and the original addresses it replaced. Cookies marked HttpOnly stay out of its reach. In the recovered browser payloads, Hudson Rock found no seed phrase extraction and no wallet signing calls.
The fourth module targets Apple devices. On an iPhone or iPad, the final builds load router scripts from two .icu hosts, read the iOS version, and open an exploit chain in a hidden frame. Report URI's timeline shows the threat actor adding the second host for all iOS visits between 17:13 and 17:57 UTC, with a Coruna frame on every visit from iOS 18.6 or earlier beside the version-routed exploit chains.
The router sends iOS 13 to 17 to Coruna and iOS 18 to DarkSword, and a separate route for iOS 26 only fingerprinted devices when Report URI checked. The kit's own comments list iOS 18.7.3 and later, and 26.3 and later, as patched.
Report URI counts 26 rebuilds of the payload in one day, with the threat actor adding features and wallets but never removing a wallet. StepSecurity found the payload server set a 15-second cache lifetime, letting each new build reach visitors within seconds. The final build went live at 17:56 UTC by its own Last-Modified header, and Report URI dates it to 17:57 UTC. Matching on one file hash misses earlier builds, so defenders need the domains and behavior rather than a single indicator.
Crypto Receipts and an Unknown Remainder
The wallets had received about US$7,040 by 19:25 UTC on November 16, and by my sum the receipts in Report URI's November 17 update total about US$7,150. Most arrived in Bitcoin, the rest in Litecoin, Ethereum, TRON, and stablecoins on TRON, Ethereum, and BNB Smart Chain. The threat actor swept part of the funds from 13:03 to 13:05 UTC and emptied the wallets between 04:11 and 05:39 UTC on November 17. StepSecurity found a further 1.7065 litecoin (LTC), about US$75, arriving at about 10:18 UTC.
Galaxy Digital's Alex Thorn, cited by StepSecurity, puts the victims' losses at about US$2,300, reading most of the Bitcoin as the threat actor's own funds routed back through an exchange. StepSecurity could not confirm that reading and notes the batched Bitcoin transfers fit either explanation.
At least two of the wallets predate the Optimizely injection. The TRON wallet has been in use since March 2026, and its last 116 Tether (USDT) went to an address also receiving cash-outs on March 24 and November 3, according to Report URI and StepSecurity. StepSecurity found the Ethereum wallet received 0.0192 ether (ETH) on November 12, four days before 1688360[.]cc existed. Report URI ties the cash-out address to earlier attacks.
Blockchain receipts measure only one of four modules. No public source shows how many cards reached card.php, how many cookies and screenshots reached hit.php, or whether any iPhone was compromised.
Public scans presenting iOS user-agent strings show the router serving Coruna and DarkSword entry pages, but Hudson Rock notes those scans run in headless Chrome, which cannot reproduce Safari's WebKit engine or an iOS kernel. Hudson Rock concludes the evidence connects the campaign to exploit-kit delivery infrastructure. It says the evidence does not establish a zero-day or successful device exploitation.
Hudson Rock's later analysis reports researcher Kirk of Ohey retrieving further DarkSword stages, including post_exploit.js, built to register the device and upload its keychain and other databases. Ohey also describes panel features meant to collect seed phrases and private keys. Those findings describe what the later stages were designed to do, beyond the browser payload's theft of addresses, cards, and cookies.
From Spy Tool to 388 USDT
Google Threat Intelligence Group (GTIG) disclosed Coruna on March 4, 2026, describing five full exploit chains and 23 exploits against iOS 13 through 17.2.1. GTIG first captured parts of it in February 2025 from a customer of a surveillance company. UNC6353, a suspected Russian espionage threat actor, then used it in watering hole attacks on Ukrainian users. UNC6691, a financially motivated threat actor operating from China, later deployed it on fake Chinese finance sites.
GTIG called Coruna's path unclear but said it suggests an active market for second-hand zero-day exploits. Its DarkSword analysis on March 19 described a full exploit chain for iOS 18.4 through 18.7 built on six vulnerabilities. Since at least November 2025, commercial surveillance vendors and suspected state-sponsored actors have used it against targets in Saudi Arabia, Turkey, Malaysia, and Ukraine. Turkey's PARS Defense was one vendor, and UNC6353 added DarkSword to its watering holes too.
Days later, someone published a newer DarkSword build on GitHub. iVerify co-founder Matthias Frielingsdorf told TechCrunch the exploits work out of the box and require no iOS expertise. Apple says devices running current software are not at risk, and it had already issued an emergency update on March 11 for devices unable to run recent iOS versions. TechCrunch, citing Apple's own data on out-of-date devices, put the exposed population at likely hundreds of millions.
By autumn, the two kits had merged into a product. Kirk's Ohey research found the combined kit sold on two Chinese source-code markets as "Caishen iOS C2 V2." One listing names Coruna, DarkSword, and an iOS 26 route as its three exploit chains and showed five sales when Ohey read it. A November 4 listing on the other market priced it at 388 USDT. When DarkSword fails on a phone, the kit falls back to Coruna, and both report to the same server.
Ohey traced the kit's lighter router across 72 sites between October 4 and November 5, on domains imitating WhatsApp, Coupang, Hepsiburada, and Mizuho Bank. The router's comments name its builder as Telegram user @caishenge, offering setup, custom development, and site maintenance. Report URI found the Coruna loader served through Optimizely byte-identical to a sample from a different campaign it analyzed in October.
Ohey found no working exploit behind the iOS 26 route. The route was switched off on a related kit server when Ohey checked it on November 5. By November 16, a router from the same kit, carried to major sites by the compromised script vendors, sent iOS 26 visits to the route again.
In an update at 22:30 UTC that night, Ohey reported an iOS 26.3 visit reaching it and loading rce_worker_26.x.js, a file whose header calls it a "GhostWave 26.x lab fork." Ohey identifies GhostWave as a DarkSword fork, and the exploit chain's own log read usable_grade=DEAD weaponized=false.
Kirk's own earlier test, reported by Hudson Rock, presented iOS 26.3 and never loaded track.js. Hudson Rock also notes Safari 26 reporting an older iOS version in its user-agent string, so the router's version labels may not match what real devices receive.
My earlier analysis of COZY BEAR (aka Midnight Blizzard) iPhone attacks found the Russian threat actor reusing DarkSword's vulnerabilities against phones the patch cycle leaves behind. Sekoia, attributing that campaign to COZY BEAR with moderate to high confidence, found it running the DarkSword exploit chain and reimplementing its CVE-2025-43529 exploit in a second exploit chain.
Four Vendors Sharing One Router
Hudson Rock and Kirk traced poisoned responses from three providers besides Optimizely, analytics service Kissmetrics, story widget StorifyMe, and video loader minute-ly. The earliest capture Hudson Rock published shows a poisoned Kissmetrics script on the site of Security Bank, a Philippine bank, at 05:42 UTC on November 13. A poisoned StorifyMe widget appeared on the German newspaper Frankfurter Allgemeine Zeitung that evening, and minute-ly's loader carried the code on Fox Sports on November 16.
These three vendors loaded webshell[.]lol/get.js, a small loader leading to the same iOS router. Both the webshell[.]lol path and the Optimizely payload reference the same channel code, 02111110. Report URI dates webshell[.]lol to September 17, when it first appeared on Turkish websites through a search engine optimization backlink panel. Public scans cited by StepSecurity show the router host 5vobauua3ojpklx[.]icu serving /A8/a.html on November 12, four days before the Optimizely wave, and Report URI identifies /A8/ as the Coruna path.
Hudson Rock says the shared hosts, loader code, and routing marker support a technical relationship between the deliveries. In its view they do not identify one threat actor or prove a common way into all four vendors. A commercial kit can serve many buyers, as Hudson Rock notes. On the published dates, the Optimizely wave was the widest known use of that infrastructure.
Trust placed in one supplier has carried other recent attacks too. Expedia said this month threat actors reached its systems through a compromised third-party data platform provider, though it has not named the provider. In August, one hijacked maintainer account put 245 million Rust downloads at risk.
Who Ran OPTIMJACK
Optimizely says attribution is pending, and no researcher has tied the campaign to a known threat actor. Its interim report names infrastructure absent from the researchers' findings. The report, citing Mandiant, names pay.pay-cloud[.]online as a host for the skimmer and lists pay-relay.wordpressadmin001.workers[.]dev as the skimmer's exfiltration endpoint. Report URI instead traces the payload and its collection endpoints to 1688360[.]cc.
The published evidence does not show what the two sets mean. Optimizely may be describing another stage of the same payload, since its account of the script swapping wallet addresses, sending a screenshot with document.cookie, and attempting an iOS exploit kit install matches the track.js modules Report URI documented. The difference could also mean more than one party used the same access, each loading its own code, which would fit the separate access broker hypothesis below.
Optimizely's preliminary research says the threat actor has run two command-and-control frameworks, ADAPTAGENT and CUPCAKE C2, from pay-cloud[.]online since at least September, and the same infrastructure hosts JavaScript card skimmers used in campaigns against other victims. The published evidence rules out none of four hypotheses.
Caishen Kit Buyer
A financially motivated threat actor bought or rented the Caishen kit and paired it with its own clipper and skimmer. The payload's design favors theft, the TRON wallet shows cash-outs since March, and the kit is openly for sale with setup services. GTIG already documented UNC6691 running Coruna at scale on fake finance sites.
Optimizely's account of that infrastructure supports this reading. It shows a theft business running before the Optimizely wave, on infrastructure apart from the kit's routers, fitting a buyer adding its own tools to a purchased kit.
Kit identifiers JIUJIU888 and __DAILI_CFG are Chinese pinyin. Jiujiu can read as 九九, "nine nine," or 久久, "for a long time." The 888 suffix repeats a number Chinese culture counts as lucky for its sound, "to prosper," matching the kit's name, Caishen, the Chinese god of wealth.
Daili (代理) means both "agent" and "proxy." Hudson Rock found the router's hidden exploit frame carrying a data-daili-gate attribute and its loader a channel code, markings fitting a reseller scheme as well as a network proxy. The language points to Chinese-speaking developers, matching the Chinese markets selling the kit, but says nothing about who bought it.
Kit Builder Running Its Own Campaign
The developer behind @caishenge could run campaigns directly, using its own product as a toolkit. Router builds carry several different Telegram handles, suggesting more than one party customizes the kit, and nothing public separates the seller from the users. The 26 payload rebuilds point to a threat actor developing its own theft code. They do not show whether that threat actor also wrote the exploit kit. Optimizely's evidence weakens this reading only slightly, since a kit builder could also run its own skimmer infrastructure.
Espionage Behind a Criminal Front
UNC6353 is the only threat actor GTIG's two March write-ups name as using both kits, and GTIG ties it to suspected Russian espionage but not to COZY BEAR. The COZY BEAR campaign Sekoia documented reached visitors from August 27 to September 3 through Plerdy SEO Tracker, a compromised analytics script, much like the poisoned Kissmetrics script. A state actor could hide iPhone targeting inside a noisy crypto theft campaign to blur attribution.
Against this reading, the campaign reloaded older iPhones' pages every 60 seconds, exposed its infrastructure within hours, used publicly sold exploits, and swept small crypto sums. UNC6353's watering holes aimed at Ukrainian users, and most sites carrying the poisoned Plerdy tracker were Ukrainian, while OPTIMJACK reached anyone loading the affected pages. Optimizely's account of that infrastructure weakens this reading further, since it fits an established eCrime business.
Separate Access Broker
Someone else could have found the Optimizely sign-up flaw and sold write access to the threat actor. Initial access brokers sell this kind of entry. A broker also fits the Smartsheet reinfection, since continued access says nothing about who first found the flaw, and the earlier vendor compromises could have come from different sources.
Optimizely describes one threat actor creating the accounts, guessing its way into customer accounts, and editing the code, which weakens this reading. Its report links that activity only to IP addresses including commercial VPN addresses, so a broker handing over working access remains possible, and the two infrastructure sets leave room for more than one party.
The evidence favors financial motivation. The payload targets the money and payment data of everyone it reaches, the exploit kit is a commercial product with a public price, and Optimizely's account of that infrastructure points to an existing theft business. I assess with moderate confidence a financially motivated threat actor using the commercially sold Caishen kit ran OPTIMJACK. Optimizely's evidence strengthens it without raising it, since the evidence is preliminary, single-source, and tied to infrastructure no researcher has matched to track.js.
I assess with low confidence the threat actor is a buyer of the kit rather than its builder, given router builds carrying several different Telegram handles and skimmer infrastructure separate from the kit's. A builder could run both, so the confidence stays low. A purchase record from either market would settle it.
Why Content Security Policy Mostly Failed
Report URI crawled 256 hosts belonging to affected organizations to test whether Content Security Policy (CSP), the browser control limiting where a page loads scripts and sends data, would have stopped the payload. Among the 159 organizations in its results, by my count, 68 had no policy at all, and 50 had a policy allowing both the payload and the data theft. Seventeen let the payload run but blocked its data exfiltration, and 24, about 15 percent, blocked it outright.
Report URI sells CSP monitoring, so these figures argue for its own product. It published its method and counts, and its point about strict-dynamic stands on Optimizely's own guidance. Allowing cdn.optimizely.com protected nobody, because the malicious code arrived from that host. The sites with a restrictive script-src list, naming only the hosts they need, blocked 1688360[.]cc.
A policy using strict-dynamic defeated that protection by trusting any script a trusted script inserts, and unsafe-eval let the inline fallback run even where the remote file was blocked. Optimizely's own CSP guidance puts strict-dynamic in its recommended policy. Report URI says Optimizely's documentation also tells customers to allow unsafe-eval, without naming the page, and the CSP guidance linked here does not mention it. At HSBC, the policy blocked track.js and the data theft, but the inline copy still ran through eval.
StepSecurity also finds Subresource Integrity unable to pin a snippet changing every time a team publishes an experiment. A consent banner decides only when the snippet runs, never what it contains, and on GitLab's site the consent manager itself loaded the Optimizely file.
Since March 31, 2025, Payment Card Industry Data Security Standard (PCI DSS) Requirements 6.4.3 and 11.6.1 have required merchants to authorize every payment page script, check its integrity, and monitor the page for tampering. Any affected site loading the snippet on a page collecting card numbers in its own form fields ran an unauthorized script there for hours. Optimizely says the injection also bypassed the protections on cdn-pci.optimizely.com, the host for its high-security PCI script.
Whether each site met those requirements is a question for its acquirer or the payment brands, since the PCI Security Standards Council does not enforce compliance itself. Palo Alto Networks' Unit 42 argues for segmentation, removing non-essential marketing, experimentation, and analytics tags from log-in and checkout pages and collecting payments only inside hardened cross-origin frames. The clipper still worked on any page displaying an address, and the iPhone branch needed no form at all.
Salesforce says it disabled Optimizely scripts across preview and production instances while it investigates. Zoom said in a statement, quoted by ABC News, it disabled Optimizely across its production environments and has found no impact to customer data, and IKEA told ABC News it is aware of the reports.
Equifax told Reuters it stopped using the platform and has found no evidence of any compromise to its systems. HSBC told Reuters on November 17 the unauthorized scripts were not active on its websites, after Report URI had found the inline copy running on an HSBC site on November 16. Most other affected organizations have said nothing publicly.
Report URI advises treating cards typed into affected sites' own forms on November 16 as potentially exposed. Hudson Rock found the embedded fallback copy carries a byte-identical skimmer, and no public source dates when that copy entered the snippets.
Optimizely tells customers credentials and API tokens do not need to be reset, which follows from its finding that the flaw bypassed authentication. Its findings are interim, so the review Report URI and StepSecurity both recommend comes first, covering every user, API token, and integration able to change a project. Report URI also says to rotate credentials, and any credential the review cannot account for should be rotated.
Report URI advises treating session tokens in cookies not marked HttpOnly as exposed and rotating them. Hudson Rock tells sites loading the Kissmetrics, StorifyMe, or minute-ly scripts to check what those scripts served as well.
Anyone who copied a wallet address or sent cryptocurrency on or after November 16, from an affected page or from anywhere else while one stayed open, should check where the funds went, since the wallet receipts are the only measurable sign of loss so far.
Anyone who typed a card into an affected site's own form on November 16 should watch the statement or ask the issuer for a new card, and anyone signed in to a sensitive service there should watch the account for activity they did not make. iPhones and iPads need the latest iOS release Apple offers them.
Those steps fall to visitors because at least 178 organizations let one vendor's script change their pages, and one sign-up flaw at that vendor turned their sites into delivery points for theft and iPhone exploits within a day.
Analyst Annex
Evidence status as of November 18, 2026. The records cover OPTIMJACK, the malicious JavaScript injected into Optimizely Web Experimentation snippets on November 16, 2026, and the Kissmetrics, StorifyMe, and minute-ly script injections beginning November 13 sharing its iOS router infrastructure. The two stay separate record sets, because public evidence connects their infrastructure without proving one threat actor ran both.
Coruna and DarkSword use by UNC6353, UNC6691, and commercial surveillance vendors appears under Related Activity as context only. None of those threat actors has been tied to OPTIMJACK.
Confirmed marks evidence published by a victim, vendor, regulator, or government authority about its own systems, decisions, or investigation. Reported marks technical findings or attribution assessments published by a third party. Claimed marks statements by a seller or self-identified threat actor that remain unverified. Provisional marks an analytical mapping supported by reported behavior but not made by the original source.
Indicators
The table is capped at twelve network records, two hashes, and four other markers. Report URI also published six fallback and sibling .icu domains, a Coruna telemetry domain, a search engine optimization panel domain, and wallets for Solana, TON, Litecoin, Bitcoin Cash, and Dogecoin. The Report URI indicator list and the StepSecurity indicator list hold the complete sets. The legitimate cdn.optimizely.com host is omitted, because it served the payload as a trusted vendor.
| Indicator | Context | Status |
|---|---|---|
| IP Addresses | ||
136.113.106[.]136 |
Address of 1688360[.]cc, the payload and data-collection server
Source: Report URI
|
Reported |
34.96.213[.]201 |
Google Cloud address in Hong Kong hosting 5vobauua3ojpklx[.]icu and two sibling .icu domains registered on November 9
Source: StepSecurity
|
Reported |
34.87.36[.]58 |
Google Cloud address in Singapore hosting eil786npto9u5bq[.]icu and two sibling domains registered at 16:38 UTC on November 16
Source: StepSecurity
|
Reported |
| Domains | ||
1688360[.]cc |
Command-and-control server serving track.js and coded as the collection point for cards, cookies, and screenshots. Registered at 05:23 UTC on November 16 and put on hold at 00:37 UTC on November 17
Source: Report URI
|
Reported |
webshell[.]lol |
Injector serving get.js in the earlier wave, first seen on Turkish websites on September 17 and put on hold at 03:37 UTC on November 17
Source: Report URI
|
Reported |
5vobauua3ojpklx[.]icu |
First iOS router, serving boot.js. Public scans show it serving the Coruna entry path /A8/a.html on November 12
Source: Report URI, StepSecurity
|
Reported |
pay-cloud[.]online |
Threat actor infrastructure Optimizely says hosted the skimmer, with its report citing Mandiant on pay.pay-cloud[.]online as a skimmer host. Optimizely says ADAPTAGENT and CUPCAKE C2 have run from it since at least September
Source: Optimizely
|
Confirmed |
pay-relay.wordpressadmin001.workers[.]dev |
Exfiltration endpoint Optimizely names for the skimmer. Any browser request to it is a strong sign the skimmer ran, according to Optimizely Source: Optimizely | Confirmed |
eil786npto9u5bq[.]icu |
Second iOS router, added between 17:13 and 17:57 UTC on November 16, serving Coruna and iOS 26 pages Source: Report URI | Reported |
| URLs | ||
1688360[.]cc/track.js |
Main payload, rebuilt 26 times on November 16 and served with a 15-second cache lifetime Source: Report URI, StepSecurity | Reported |
1688360[.]cc/card.php |
Collection endpoint the skimmer is coded to send card numbers, expiry dates, and security codes to. Its sibling hit.php is coded to receive screenshots and cookies. No public source shows any submission arriving
Source: Report URI
|
Reported |
webshell[.]lol/get.js |
Loader delivered through poisoned Kissmetrics, StorifyMe, and minute-ly responses, leading to the same iOS router Source: Hudson Rock | Reported |
| SHA-256 Hashes | ||
f7b84b9a94f749f8ced3f0112d0a66894e8bbc3b4aa5592021d1b8976a9bad40 |
SHA-256 of the final track.js build, which Report URI dates to 17:57 UTC. Earlier builds hash differently, so match on domains and behavior
Source: Report URI
|
Reported |
814a963c0da480e2449d9b6664480156e1cd6d731014117401bf635badec3a0f |
SHA-256 of the 17:13 UTC track.js build
Source: Report URI
|
Reported |
| Script Markers and Wallets | ||
data-channel-code="02111110" |
Channel code on the iOS loader, shared by the Optimizely payload and the webshell[.]lol path
Source: Report URI, Hudson Rock
|
Reported |
19g9rZcPeiz8rUvyWpTRyXXhPnjsWrc3kW |
Bitcoin wallet receiving 0.08528 BTC in five payments, about US$5,489 Source: Report URI | Reported |
THvs6p56XkUiMkJeSc51cAP64u2RsTYqxa |
TRON wallet in use since March 2026, receiving 662.9 USDT and 183.7 TRX Source: Report URI, StepSecurity | Reported |
0xC4026b06D82b0F0B363eda0642D61CfbDc17884e |
Ethereum and BNB Smart Chain wallet, receiving 0.0192 ETH on November 12, before 1688360[.]cc existed
Source: StepSecurity
|
Reported |
MITRE ATT&CK Mapping
| Tactic | Technique | Procedure | Status |
|---|---|---|---|
| OPTIMJACK Through Optimizely | |||
| Reconnaissance | T1592.002 Gather Victim Host Information: Software | The iOS routers read the visiting device’s iOS version from the browser to choose Coruna, DarkSword, or the iOS 26 route Source: Report URI | Provisional |
| Resource Development | T1583.001 Acquire Infrastructure: Domains |
Registered 1688360[.]cc at 05:23 UTC on November 16
Source: Report URI, StepSecurity
|
Reported |
| Resource Development | T1588.005 Obtain Capabilities: Exploits | Used the Coruna and DarkSword exploit kits, sold as Caishen iOS C2 V2 on two Chinese markets, one listing at 388 USDT on November 4. Whether the threat actor bought the kit or built it is not established Source: Ohey | Provisional |
| Initial Access | T1190 Exploit Public-Facing Application | Exploited a flaw in the Optimizely Web Experimentation account sign-up flow, letting created accounts act as existing users on customer accounts through a combination of public information and brute force, then edited Project JavaScript as administrator users Source: BleepingComputer, Optimizely | Confirmed |
| Initial Access | T1195.002 Supply Chain Compromise: Compromise Software Supply Chain |
Wrote a loader into the Project JavaScript field of customer projects, so cdn.optimizely.com served it to visitors of 178 organizations under Optimizely’s certificate
Source: Report URI
|
Reported |
| Initial Access | T1189 Drive-by Compromise | Visitors received the payload by loading an affected page, with no action required Source: Report URI | Reported |
| Execution | T1059.007 Command and Scripting Interpreter: JavaScript |
track.js ran in visitors’ browsers. An inline copy ran through eval when the remote file errored or did not load within 2.5 seconds
Source: Report URI, Hudson Rock
|
Reported |
| Execution | T1203 Exploitation for Client Execution | Opened Coruna and DarkSword entry pages in a hidden frame on iPhones and iPads. Hudson Rock found the evidence does not establish successful exploitation of any device Source: Report URI, Hudson Rock | Provisional |
| Stealth | T1027 Obfuscated Files or Information |
After cleanup, the Smartsheet snippet was reinfected with a loader hiding the payload URL in Base64, defeating searches for 1688360
Source: Report URI
|
Reported |
| Credential Access | T1539 Steal Web Session Cookie |
Coded to send up to 8,000 characters of cookies not marked HttpOnly to hit.php after a wallet swap
Source: StepSecurity
|
Reported |
| Collection | T1056.003 Input Capture: Web Portal Capture | Watched form submissions and outgoing requests for a Luhn-valid card number, expiry date, and security code Source: Report URI, StepSecurity | Provisional |
| Collection | T1113 Screen Capture |
Coded to render the visible page with html2canvas and post the image to hit.php
Source: Report URI
|
Reported |
| Command and Control | T1071.001 Application Layer Protocol: Web Protocols | Rebuilt the payload 26 times, served over HTTPS with a 15-second cache lifetime, so each reached visitors within seconds Source: Report URI, StepSecurity | Reported |
| Exfiltration | T1041 Exfiltration Over C2 Channel |
Coded to beacon card data to card.php and screenshots, cookies, and replaced addresses to hit.php on 1688360[.]cc
Source: Report URI
|
Reported |
| Impact | T1565.002 Data Manipulation: Transmitted Data Manipulation |
Replaced wallet addresses for eight cryptocurrencies in page text, form fields, links, the clipboard, outgoing requests, and the iOS share sheet, hooking Function.prototype.call and apply
Source: Report URI, StepSecurity
|
Reported |
| Other Script Vendor Injections | |||
| Initial Access | T1195.002 Supply Chain Compromise: Compromise Software Supply Chain |
Poisoned Kissmetrics, StorifyMe, and minute-ly responses loaded webshell[.]lol/get.js, with the earliest published capture on Security Bank’s site at 05:42 UTC on November 13. How each vendor’s scripts were altered is not established
Source: Hudson Rock
|
Reported |
| Initial Access | T1189 Drive-by Compromise |
webshell[.]lol first appeared on Turkish websites on September 17 through a search engine optimization backlink panel
Source: Report URI
|
Reported |
Entities
| Entity | Role | Status |
|---|---|---|
| Threat Actors | ||
| Unidentified OPTIMJACK Threat ActorThreat actor, assessed eCrime |
Ran the November 16 injection with a payload built for theft. Its TRON wallet has been in use since March 2026, and Report URI ties its cash-out address to earlier attacks. Optimizely says attribution is pending and ties the threat actor to pay-cloud[.]online, where JavaScript card skimmers used against other victims are hosted. Financial motivation is a jark.me assessment at moderate confidence, and espionage behind a criminal front remains open
Source: Report URI, StepSecurity, Optimizely
|
Reported |
@caishengeTelegram user, kit developer |
Named in router comments as the builder of the Caishen kit, offering setup, custom development, and site maintenance. Other router builds carry different handles Source: Ohey | Claimed |
| Organizations and Victims | ||
| OptimizelyBreached vendor, New York | Confirmed an unauthorized party exploited a flaw in its Web Experimentation account sign-up flow to impersonate users on targeted customer accounts and add malicious JavaScript to their projects. Says less than 5 percent of its customers were impacted and has engaged Mandiant Source: Optimizely, Optimizely | Confirmed |
| 178 Affected OrganizationsVictims | Report URI found infected snippets belonging to Zoom, the BBC, CNN, Salesforce, GitLab, IKEA, HSBC, Equifax, Air France-KLM, Ledger, and others, among the 1,611 snippets it checked Source: Report URI | Reported |
| ZoomVictim | Disabled Optimizely across its production environments and found no impact to customer data Source: ABC News | Confirmed |
| SalesforceVictim | Disabled Optimizely scripts across preview and production instances while it investigates Source: BleepingComputer | Confirmed |
| Three Other Script VendorsThird-party script providers |
Kissmetrics, StorifyMe, and minute-ly served poisoned responses carrying the webshell[.]lol loader between November 13 and November 16
Source: Hudson Rock
|
Reported |
| Tools and Services | ||
| Caishen iOS C2 V2Commercial exploit kit | Combines Coruna, DarkSword, and an iOS 26 route. Sold on two Chinese source-code markets, one listing at 388 USDT on November 4 Source: Ohey | Claimed |
| CorunaiOS exploit kit |
Five full exploit chains and 23 exploits against iOS 13 through 17.2.1. Served to iOS 13 to 17 visitors through the /A8/ path and, from a change made between 17:13 and 17:57 UTC on November 16, framed into every visit from iOS 18.6 or earlier
Source: GTIG, Report URI
|
Reported |
| DarkSwordiOS exploit chain | Built on six vulnerabilities for iOS 18.4 through 18.7. Served to iOS 18 visitors Source: GTIG, Report URI | Reported |
track.jsMalicious JavaScript payload |
Combined a crypto clipper, card skimmer, screenshot and cookie stealer, and iOS exploit loader in one file Source: Report URI, StepSecurity | Reported |
| Vulnerabilities | ||
| Optimizely Experimentation Sign-Up FlawVulnerability, no CVE published | Let accounts created through sign-up act as existing users on customer accounts and bypassed authentication. Optimizely says it closed the flaw at 16:10 UTC on November 16 and has not described the flaw itself Source: Optimizely | Confirmed |
Related Activity
The records below cover earlier use of the Coruna and DarkSword exploits, the Caishen router’s other sites, and Optimizely’s February 2026 breach. None was observed in OPTIMJACK.
| Item | Detail | Status |
|---|---|---|
| Coruna Watering HolesCampaign, 2025 | GTIG first captured parts of Coruna in February 2025 from a surveillance company’s customer. UNC6353, a suspected Russian espionage threat actor, then used it against Ukrainian users Source: GTIG | Reported |
| UNC6691 Fake Finance SitesCampaign, reported March 2026 | A financially motivated threat actor operating from China deployed Coruna on fake Chinese finance sites Source: GTIG | Reported |
| DarkSword Surveillance UseCampaign, November 2025 onward | Commercial surveillance vendors, including Turkey’s PARS Defense, and suspected state-sponsored actors used DarkSword against targets in Saudi Arabia, Turkey, Malaysia, and Ukraine Source: GTIG | Reported |
| DarkSword GitHub LeakLeak, March 2026 | Someone published a newer DarkSword build on GitHub. iVerify told TechCrunch the exploits work out of the box Source: TechCrunch | Reported |
| Caishen Router SitesCampaign, October 4 to November 5, 2026 | The kit’s lighter router ran on 72 sites imitating WhatsApp, Coupang, Hepsiburada, and Mizuho Bank Source: Ohey | Reported |
| Optimizely Voice-Phishing BreachIntrusion, disclosed February 2026 | A threat actor reached Zendesk, Salesforce CRM records, and internal documents through voice phishing. Optimizely said the threat actor could not escalate privileges Source: SecurityWeek | Confirmed |
| COZY BEAR DarkSword ChainCampaign, reported November 2026 | Sekoia found the Russian threat actor reimplementing DarkSword’s CVE-2025-43529 exploit and reusing the DarkSword exploit chain against unpatched iPhones Source: Sekoia, jark.me | Reported |
Intelligence Gaps
| Gap | Why It Matters | What Would Close It |
|---|---|---|
| Mechanism of the sign-up flaw | Optimizely says created accounts could act as existing users through public information and brute force, bypassing authentication, but has not described the flaw. Without the mechanism, other platforms cannot check for the same class of flaw | Optimizely’s final root-cause analysis after the Mandiant investigation |
| Two accounts of the timeline and infrastructure | Optimizely dates the first malicious code to 07:51 UTC and calls delivery fully blocked from 16:06 UTC, while Report URI saw the BBC snippet loading code at 04:21 UTC and found 186 infected snippets from 19:24 to 19:59 UTC. Optimizely names pay-relay.wordpressadmin001.workers[.]dev as the exfiltration endpoint, where Report URI names 1688360[.]cc |
Optimizely’s final indicators of compromise or per-snippet revision logs |
| How long the flaw existed and who else used it | Optimizely dates the first impersonated user account to 00:37 UTC on November 16 but has not said how long the vulnerability was present, so earlier abuse cannot be ruled out | Optimizely audit logs of project changes before November 16 |
| Stolen card and cookie data | Blockchain receipts measure only the clipper. No public source shows how many cards, cookies, or screenshots reached 1688360[.]cc |
Seized server data from 1688360[.]cc, card brand fraud data, or victim disclosures |
| Any iPhone compromise | Public scans run in headless Chrome and cannot show whether Coruna or DarkSword succeeded on a real device. Ohey has retrieved later DarkSword stages built to upload the keychain and other databases, but no public source shows them running on a real device | Forensic findings from an affected iPhone or Apple threat notifications tied to the campaign |
| Whether one threat actor ran all four vendor injections | Shared infrastructure connects the deliveries without proving common control or a common way in | Kissmetrics, StorifyMe, and minute-ly incident findings, or server-side evidence from the router hosts |
| Buyer or builder of the Caishen kit | Separates a customer of a commercial product from the developer selling it to others | Market records, Telegram communications, or law enforcement findings |
| Full victim count | Report URI checked only the 1,611 snippets it could find in time. Optimizely says less than 5 percent of its customers were impacted and cannot disclose its customers | An Optimizely count of affected customer projects |
| Victim losses in crypto | Galaxy Digital puts victim losses near US$2,300 against about US$7,150, the sum of Report URI’s November 17 per-wallet figures, and StepSecurity could not confirm Galaxy Digital’s reading | Exchange records tracing the Bitcoin deposits to their senders |