Threat reportMalwareTL-2026-0403
Weaponizing Trust Signals: Claude Code Lures and GitHub Release Payloads Deliver Vidar Stealer v18.7 + GhostSocks
Financially motivated threat actors weaponized the 2026-03-31 Anthropic Claude Code npm packaging incident by standing up trojanized GitHub repositories branded 'leaked-claude-code' within 24 hours. The distribution hub github.com/leaked-claude-code/leaked-claude-code (owned by account idbzoomh1) served 7z archives containing a Rust-compiled dropper (TradeAI.exe / ClaudeCode_x64.exe) that installs Vidar Stealer v18.7 and GhostSocks SOCKS5 proxy malware.
Key facts
| Identifiers | TL-2026-0403 |
|---|---|
| Also known as | Claude Code Lure Campaign, leaked-claude-code Campaign, TradeAI Dropper Campaign, Vidar + GhostSocks GitHub Releases Campaign, idbzoomh1 Operation |
| Severity | HIGH |
| CVSS | 8.1 (CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:L) |
| Affected products | Microsoft Windows; Anthropic Claude Code (impersonated, not actually vulnerable); GitHub GitHub Releases (abused distribution channel) |
| Weaknesses (CWE) | CWE-506, CWE-807, CWE-494 |
| Malware | GhostSocks, Vidar Stealer, Vidar v18.7 |
| Tools | 7-Zip archive packaging |
| MITRE ATT&CK techniques | T1005 Data from Local System, T1016 System Network Configuration Discovery, T1027 Obfuscated Files or Information, T1027.002 Obfuscated Files or Information: Software Packing, T1036.005 Match Legitimate Resource Name or Location, T1041 Exfiltration Over C2 Channel, T1059.003 Command and Scripting Interpreter: Windows Command Shell, T1071.001 Application Layer Protocol: Web Protocols, T1082 System Information Discovery, T1083 File and Directory Discovery, T1090.002 Proxy: External Proxy, T1102.001 Web Service: Dead Drop Resolver and 21 more in the attack flow |
| Motivation | FINANCIAL |
| Target sectors | technology, software-development, cryptocurrency, financial, ai-research, general-consumer, freelance-developers, saas |
| Target regions | Global, North America, Europe, Asia-Pacific |
| Detection rules | 9 (1 shown free) |
| Sources | 9 cited sources |
Detection coverage
As of 2026-04-21, Threadlinqs Intelligence publishes 9 detection rule(s) for TL-2026-0403 across Splunk SPL, Microsoft KQL and Sigma, covering 20 indicator(s) of compromise. The whole corpus is readable without an account. Every threat comes with one complete detection in Splunk SPL, Microsoft KQL and Sigma, free for everyone; a free account adds daily TL Creds to unlock the rest. Paid plans add a larger daily allowance, correlation and the MCP server. Threadlinqs MCP server · View plans.
Free detection: Claude Code Lure — Password-Protected 7z Extraction of leaked-claude-code Archive (Exploitation)
Detects execution of 7z.exe/7zG.exe/7zFM.exe extracting an archive whose name matches the leaked-claude-code / TradeAI / ClaudeCode_x64 lure patterns with a -p password supplied on the command line. This is the initial-access primitive of the TL-2026-0403 campaign where victims download a password-protected 7z from github.com/leaked-claude-code Releases and extract using the plaintext password from the repo README. The password-on-CLI pattern defeats gateway AV and is a high-fidelity indicator of the lure.
- Languages: Splunk SPL, Microsoft KQL, Sigma
- ATT&CK: T1566.002, T1204.002, T1027.002
- Data sources: sysmon, windows_security, edr_process_creation
- Severity: high
- Confidence: high
Splunk SPL
index=sysmon sourcetype="XmlWinEventLog:Microsoft-Windows-Sysmon/Operational" EventCode=1
(Image="*\\7z.exe" OR Image="*\\7zG.exe" OR Image="*\\7zFM.exe")
(CommandLine="*claude-code-leaked*" OR CommandLine="*ClaudeCode_x64*" OR CommandLine="*TradeAI*")
CommandLine="* -p*"
| eval lure_family=case(
match(CommandLine,"(?i)claude-code-leaked"),"leaked-claude-code",
match(CommandLine,"(?i)ClaudeCode_x64"),"ClaudeCode_x64",
match(CommandLine,"(?i)TradeAI"),"TradeAI",
true(),"unknown")
| stats count min(_time) as first_seen max(_time) as last_seen values(CommandLine) as cmdlines values(ParentImage) as parents by host, user, lure_family
| table first_seen last_seen host user lure_family count cmdlines parentsMicrosoft KQL
DeviceProcessEvents
| where Timestamp > ago(24h)
| where FileName in~ ("7z.exe","7zG.exe","7zFM.exe")
| where ProcessCommandLine has_any ("claude-code-leaked","ClaudeCode_x64","TradeAI")
| where ProcessCommandLine matches regex @"\s-p\S"
| project Timestamp, DeviceName, AccountName, FileName, ProcessCommandLine, InitiatingProcessFileName, FolderPath
| extend LureFamily = case(
ProcessCommandLine has "claude-code-leaked", "leaked-claude-code",
ProcessCommandLine has "ClaudeCode_x64", "ClaudeCode_x64",
ProcessCommandLine has "TradeAI", "TradeAI",
"unknown")
| order by Timestamp descSigma
title: Claude Code Lure — Password-Protected 7z Extraction (TL-2026-0403)
id: 7a1cdb01-0403-4a12-a101-claudecode0403a
status: experimental
description: Detects 7z extraction of an archive named like claude-code-leaked*.7z / TradeAI*.7z / ClaudeCode_x64*.7z using a password supplied on the command line — the initial-access primitive of the leaked-claude-code GitHub Release lure campaign.
references:
- https://www.trendmicro.com/en_us/research/26/d/weaponizing-trust-claude-code-lures-and-github-release-payloads.html
- https://www.bleepingcomputer.com/news/security/claude-code-leak-used-to-push-infostealer-malware-on-github/
- https://attack.mitre.org/techniques/T1204/002/
author: Threadlinqs Team
date: 2026-04-21
tags:
- attack.initial_access
- attack.t1566.002
- attack.execution
- attack.t1204.002
- attack.defense_evasion
- attack.t1027.002
logsource:
product: windows
category: process_creation
detection:
selection_image:
Image|endswith:
- '\7z.exe'
- '\7zG.exe'
- '\7zFM.exe'
selection_lure:
CommandLine|contains:
- 'claude-code-leaked'
- 'ClaudeCode_x64'
- 'TradeAI'
selection_pw:
CommandLine|contains: ' -p'
condition: selection_image and selection_lure and selection_pw
falsepositives:
- Security researchers intentionally detonating labelled samples in a sandbox
- Incident responders extracting a captured archive for analysis
level: highKnown false positives:
- Security researchers detonating labelled samples in sandboxes
- Incident responders extracting captured archives during analysis
- Developers with a legitimate archive that coincidentally contains the token 'TradeAI'
Tuning notes: Exclude known DFIR/forensics workstations by hostname or account. If 7-Zip is rare in the environment, drop the lure-name filter and alert on any password-protected extraction from user-writable paths. Consider enriching with download source (SmartScreen/EDR referrer) to correlate to github.com/leaked-claude-code/*.
- Rust-Compiled Dropper Execution From User Archive Path — TradeAI/ClaudeCode_x64 (Exploitation) · unlock with TL Creds
- Vidar Stealer Credential Log Staged in User Temp Path (Post-Exploitation) · unlock with TL Creds
- Dropper HTTPS Exfiltration to Non-CDN Host After Dead-Drop Resolver Fetch (C2) · unlock with TL Creds
- GitHub Release Download From [indicator] or Known Rotating-Brand Lure Repository (Exploitation) · unlock with TL Creds
- Rust-Compiled Dropper Spawning Masquerading PowerShell/Cmd Child (Post-Exploitation) · unlock with TL Creds
- GhostSocks Residential-Proxy Beacon — Sustained SOCKS5 From Unsigned User-Land Binary (C2/Lateral Movement) · unlock with TL Creds
- Vidar Dead-Drop Resolver — Non-Browser Process Fetching Steam or Telegram Profile (Post-Exploitation) · unlock with TL Creds
- GhostSocks SOCKS5 Outbound Tunnel From Non-Browser User-Land Binary (C2) · unlock with TL Creds
Indicators of compromise
As of 2026-04-21, Threadlinqs Intelligence tracks 20 indicators of compromise for TL-2026-0403.
Technical analysis
The Claude Code lure is one rotation in a broader campaign active since February 2026 that has impersonated 25+ software brands across 38 archives with 22 payload variants, abusing GitHub Releases as a trusted malware delivery channel.
On 2026-03-31 Anthropic published a malformed Claude Code npm package that briefly exposed approximately 512,000 lines of internal TypeScript source via a 59.8 MB source map. Within 24 hours a threat actor using GitHub account 'idbzoomh1' created a public repository at github.com/leaked-claude-code/leaked-claude-code framed as a re-hosting of the 'leaked' Claude Code source. The repository README and releases lured visitors to download 7z archives that, instead of the advertised TypeScript, contained a Rust-compiled Windows PE dropper delivered under multiple filenames including TradeAI.exe and ClaudeCode_x64.exe.
Upon execution, the dropper decrypts and stages two payloads: Vidar Stealer v18.7 (a mature subscription-based infostealer) and GhostSocks (a Go-compiled SOCKS5 residential-proxy implant that converts victim machines into exit nodes for the operator's proxy service). Vidar v18.7 retains the family's signature dead-drop-resolver technique, using attacker-controlled Steam and Telegram profile pages to host obfuscated strings that resolve the current C2 endpoint, giving the operator cheap and resilient C2 rotation without DNS registration. Vidar harvests browser credentials, cookies, autofill data, cryptocurrency wallets (desktop and browser extensions), MFA/2FA tokens, FTP and email client credentials, Telegram and Discord session artifacts, screenshots, and system fingerprinting data before packaging the logs and exfiltrating them over HTTPS to the rotating C2. GhostSocks then establishes a persistent outbound SOCKS5 tunnel that the operator or downstream customer can relay traffic through, enabling follow-on fraud, account takeover, and anonymized access operations against the victim's IP reputation.
Trend Micro, Zscaler ThreatLabz, and SOC Prime independently track this Claude Code lure as one rotation in a broader rotating-brand campaign running continuously since February 2026. Trend Micro fingerprinted 25+ impersonated software brands across 38 distinct GitHub archive drops, paired with at least 22 payload variants of the Vidar + GhostSocks combo. Lures rotate to whatever news cycle favors them: trading and AI tooling ('TradeAI'), crypto wallets, VPNs, productivity apps, and now the Anthropic packaging incident. By 2026-04-07 the leaked-claude-code repository had accumulated 838 stars, 1,060 forks, and 533 confirmed payload downloads, and the operator had not been suspended by GitHub at the time of the initial Trend Micro publication.
The campaign is notable for weaponizing legitimate developer trust signals: (1) GitHub Releases are commonly whitelisted by endpoint and proxy policies; (2) the github.com TLS-fronted download URL passes most reputation checks; (3) star and fork counts are inflated by the operator's botnet to signal legitimacy; and (4) the timing of the Claude Code lure, within 24 hours of a real Anthropic incident, exploits defender and developer curiosity at its peak. Defenders should treat any 'leaked source' repository appearing immediately after a public incident as high-risk, inspect GitHub Release asset hashes against known-good artifacts, and instrument egress monitoring for the dead-drop-resolver beacons (Steam and Telegram profile fetches from non-browser processes) that are characteristic of Vidar v18.7.
MITRE ATT&CK techniques by tactic:
Collection
T1005 Data from Local System; T1113 Screen Capture; T1115 Clipboard Data
Discovery
T1016 System Network Configuration Discovery; T1082 System Information Discovery; T1083 File and Directory Discovery; T1518 Software Discovery
Defense Evasion
T1027 Obfuscated Files or Information; T1027.002 Obfuscated Files or Information: Software Packing; T1036.005 Match Legitimate Resource Name or Location; T1140 Deobfuscate/Decode Files or Information; T1497 Virtualization/Sandbox Evasion
Exfiltration
T1041 Exfiltration Over C2 Channel; T1567 Exfiltration Over Web Service
Execution
T1059.003 Command and Scripting Interpreter: Windows Command Shell; T1204.002 User Execution: Malicious File
Command and Control
T1071.001 Application Layer Protocol: Web Protocols; T1090.002 Proxy: External Proxy; T1102.001 Web Service: Dead Drop Resolver; T1105 Ingress Tool Transfer; T1573.002 Encrypted Channel: Asymmetric Cryptography
Initial Access
T1189 Drive-by Compromise; T1195.002 Supply Chain Compromise: Compromise Software Supply Chain; T1566.002 Phishing: Spearphishing Link
Credential Access
T1539 Steal Web Session Cookie; T1552.001 Unsecured Credentials: Credentials In Files; T1555 Credentials from Password Stores; T1555.003 Credentials from Password Stores: Credentials from Web Browsers
Resource Development
T1583.006 Acquire Infrastructure: Web Services; T1585.001 Establish Accounts: Social Media Accounts; T1587.001 Develop Capabilities: Malware; T1588.001 Obtain Capabilities: Malware
Impact
Remediation
Patches
- No vendor patch applies: this is a malware distribution campaign, not a software vulnerability
- Anthropic revoked the malformed Claude Code npm artifact and re-published a clean package on 2026-03-31 — ensure all Claude Code installations are pulled from npm registry, not mirrored 'leaks'
Immediate actions
- Block downloads from github.com/leaked-claude-code/* at web proxy and EDR
- Hunt execution of TradeAI.exe and ClaudeCode_x64.exe across the fleet and isolate any hosts that executed them
- Rotate all credentials, session cookies, MFA tokens, and cryptocurrency wallet keys on any host that executed either dropper
- Block outbound HTTPS from non-browser processes to steamcommunity.com and t.me profile pages used as Vidar dead-drop resolvers
- Quarantine any 7z archive with filenames matching claude-code-leaked*.7z, TradeAI*.7z, or ClaudeCode_x64*.7z at mail and web gateways
Workarounds
- Filter GitHub Release downloads through a code-signing and hash-reputation gateway before execution
- Use application allowlisting (WDAC, AppLocker) to block execution of unsigned user-land binaries
Longer-term hardening
- Deploy EDR with behavioral detection for Rust-compiled droppers and dead-drop-resolver beaconing from non-browser processes
- Restrict developer workstations from installing unsigned PE executables extracted from GitHub Release 7z archives
- Add Vidar and GhostSocks YARA rules to endpoint scanners and email attachment sandboxing
- Instrument DNS and HTTPS egress telemetry for Steam profile and Telegram t.me profile lookups from non-browser parent processes
- Educate developers that 'leaked source' repositories appearing immediately after public incidents are a recurring malware lure pattern
- Enroll developer laptops in attack-surface reduction rules that block execution from user-writable archive extraction paths
Affected and fixed versions
- Microsoft — Windows
Vulnerable versions: 10; 11; Server 2019; Server 2022 - Anthropic — Claude Code (impersonated, not actually vulnerable)
Vulnerable versions: n/a — brand abuse only - GitHub — GitHub Releases (abused distribution channel)
Vulnerable versions: n/a — platform abuse
Timeline
- Rotating-lure campaign using GitHub Releases begins; first impersonated brands include trading and crypto wallet apps. Vidar Stealer + GhostSocks payload pairing observed.
- Campaign expands to multiple software brand impersonations. Trend Micro begins tracking as a unified operation linked by payload variants and GitHub account reuse patterns.
- Anthropic publishes a malformed Claude Code npm package that exposes approximately 512,000 lines of internal TypeScript via a 59.8 MB source map. The bad artifact is withdrawn the same day.
- First trojanized 7z archive drops locked-ioc-5 and locked-ioc-4 Rust-compiled droppers that stage Vidar Stealer v18.7 and GhostSocks SOCKS5 proxy.
- GitHub account 'idbzoomh1' creates repository github.com/leaked-claude-code/leaked-claude-code, pitched as a re-host of the leaked Claude Code source. Initial GitHub Release uploaded containing trojanized 7z archive.
- BleepingComputer and The Register publish the first public exposure of the trojanized Claude Code lure.
- Trend Micro, Zscaler ThreatLabz, and Help Net Security publish detailed analyses confirming Vidar v18.7 and GhostSocks payloads and linking the lure to the broader rotating-brand campaign.
- SOC Prime releases hunt queries and Sigma rules for the Claude Code lure variant and the Vidar + GhostSocks payload chain.
- leaked-claude-code repository recorded at 838 stars, 1,060 forks, and 533 confirmed payload downloads, with star/fork counts believed to be partially inflated by operator-controlled accounts.
- Trend Micro publishes follow-up noting the Claude Code lure remains live weeks after initial disclosure; the operator rotates archive filenames and release tags to evade GitHub abuse takedowns.
- Threadlinqs Intelligence publishes TL-2026-0403 with full MITRE mapping, IOC set, detection content, and attack simulation.
- As of 2026-05-29, the specific Claude Code lure rotation has cooled as a news hook, but the broader rotating-brand campaign (active since Feb 2026, 25+ brands) and its tooling persist: Vidar is 2026's top infostealer and GhostSocks-via-GitHub abuse continues. The idbzoomh1 operator is unattributed with no arrest, takedown, or sinkhole; no CVE/patch applies.
Sources cited
- Weaponizing Trust Signals: Claude Code Lures and GitHub Release Payloads Trend Micro · Research · 2026-04-03
- Claude Code Packaging Error Remains a Lure in an Active Campaign Trend Micro · Research · 2026-04-10
- Anthropic Claude Code Leak - Zscaler ThreatLabz zscaler.com · Research · 2026-04-03
- Claude Code Lures: GitHub Releases Drop Vidar & GhostSocks socprime.com · Research · 2026-04-05
- MITRE ATT&CK: T1102.001 Web Service: Dead Drop Resolver MITRE · Research · 2024-10-15
- MITRE ATT&CK: T1195.002 Supply Chain Compromise: Software Supply Chain MITRE · Research · 2024-10-15
- Claude Code leak used to push infostealer malware on GitHub BleepingComputer · News · 2026-04-02
- Fake Claude Code source downloads actually delivered malware The Register · News · 2026-04-02
- Claude Code leak GitHub malware (Help Net Security) Help Net Security · News · 2026-04-03
Previous and next in malware
- NGate Android Malware — HandyPay-Trojanized NFC Relay Variant Targets Brazilian Cardholders via Fake Rio de Prêmios Lottery and Counterfeit Google Play Lures Previous report
- Fake Google Antigravity Installer Drops NWHStealer — Credential, Session, and Crypto Wallet Theft via Trojanized AI IDE Next report
Drafted by the Threadlinqs research pipeline from 9 sources; published ; last updated —.