Skip to navigation

Threat reportMalwareTL-2026-2916

Atomic macOS (AMOS) Stealer Delivered via Malicious Ad Impersonating Claude Code (ClickFix-style)

highACTIVE

A malicious ad impersonating Claude Code sends victims to a fake Claude Code site that shows ClickFix-style instructions to paste a script into Terminal. The script installs Atomic macOS (AMOS) Stealer, which then prompts for the user's password and permissions.

Key facts

IdentifiersTL-2026-2916
Also known asAtomic Stealer, AMOS, Fake Claude Code ClickFix malvertising
SeverityHIGH
Affected productsApple macOS
MalwareAMOS, MacSync Stealer
MITRE ATT&CK techniquesT1005 Data from Local System, T1036 Masquerading, T1056.002 Input Capture: GUI Input Capture, T1059.002 Command and Scripting Interpreter: AppleScript, T1059.004 Command and Scripting Interpreter: Unix Shell, T1140 Deobfuscate/Decode Files or Information, T1204.004 User Execution: Malicious Copy and Paste, T1497.001 Virtualization/Sandbox Evasion: System Checks, T1539 Steal Web Session Cookie, T1553.001 Subvert Trust Controls: Gatekeeper Bypass, T1555.001 Credentials from Password Stores: Keychain, T1555.003 Credentials from Password Stores: Credentials from Web Browsers and 2 more in the attack flow
MotivationFINANCIAL
Target sectorstechnology, software-development, cryptocurrency, general-consumer
Target regionsGlobal
Detection rules9 (1 shown free)
Sources10 cited sources

Detection coverage

As of 2026-10-02, Threadlinqs Intelligence publishes 9 detection rule(s) for TL-2026-2916 across Splunk SPL, Microsoft KQL and Sigma, covering 13 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: AMOS fake password prompt — osascript hidden-answer dialog and dscl authonly

Detects the credential-harvesting step of AMOS and the MacSync variant. The stealer shows a fake system password dialog through osascript (display dialog with hidden answer) and validates the typed password with dscl authonly. Either signal alone is high-fidelity; both on the same host within 10 minutes is treated as critical. The search matches the post-execution behaviour that follows the ClickFix paste step.

  • Languages: Splunk SPL, Microsoft KQL, Sigma
  • ATT&CK: T1056.002, T1059.002
  • Data sources: crowdstrike:events:sensor ProcessRollup2 (macOS), DeviceProcessEvents (MDE for macOS), Sigma process_creation (macos)
  • Severity: high
  • Confidence: high

Splunk SPL

index=edr sourcetype="crowdstrike:events:sensor" event_simpleName=ProcessRollup2 event_platform=Mac earliest=-24h
| eval proc=lower(mvindex(split(ImageFileName,"/"),-1))
| eval fake_prompt=if(proc="osascript" AND match(CommandLine,"(?i)display dialog") AND match(CommandLine,"(?i)hidden answer"),1,0)
| eval pw_lure=if(fake_prompt=1 AND match(CommandLine,"(?i)(password|system (preferences|settings)|keychain|unlock|required)"),1,0)
| eval auth_check=if(proc="dscl" AND match(CommandLine,"(?i)authonly"),1,0)
| where fake_prompt=1 OR auth_check=1
| bin _time span=10m
| stats max(fake_prompt) AS fake_prompt max(pw_lure) AS pw_lure max(auth_check) AS auth_check values(CommandLine) AS cmds values(ParentBaseFileName) AS parents BY aid ComputerName UserName _time
| eval risk=if(fake_prompt=1 AND auth_check=1,"critical","high")
| table _time ComputerName UserName risk fake_prompt pw_lure auth_check parents cmds

Microsoft KQL

DeviceProcessEvents
| where Timestamp > ago(24h)
| where FolderPath startswith '/'
| extend FakePrompt = FileName =~ 'osascript' and ProcessCommandLine has 'display dialog' and ProcessCommandLine has 'hidden answer'
| extend AuthCheck = FileName =~ 'dscl' and ProcessCommandLine has 'authonly'
| where FakePrompt or AuthCheck
| summarize FakePromptHit = max(toint(FakePrompt)), AuthCheckHit = max(toint(AuthCheck)), Cmds = make_set(ProcessCommandLine, 10), Parents = make_set(InitiatingProcessFileName, 5), FirstSeen = min(Timestamp) by DeviceId, DeviceName, AccountName, WindowStart = bin(Timestamp, 10m)
| extend Risk = iff(FakePromptHit == 1 and AuthCheckHit == 1, 'critical', 'high')
| project FirstSeen, DeviceName, AccountName, Risk, FakePromptHit, AuthCheckHit, Parents, Cmds

Sigma

title: macOS Fake Password Prompt via osascript or dscl authonly Password Validation
id: 2b9c41f8-6d07-4a35-8e1c-5f3a7d0b9c02
status: experimental
description: Detects osascript showing a hidden-answer dialog (fake system password prompt) or dscl authonly being used to validate a captured password. Both behaviours are characteristic of Atomic macOS Stealer (AMOS) and the MacSync variant.
references:
    - https://www.malware-traffic-analysis.net/2026/10/02/index.html
    - https://moonlock.com/atomic-macos-stealer
    - https://www.sophos.com/en-us/blog/why-amos-matters-the-macos-malware-stealing-data-at-scale
author: Threadlinqs Team
date: 2026/10/05
tags:
    - attack.credential-access
    - attack.t1056.002
    - attack.execution
    - attack.t1059.002
logsource:
    category: process_creation
    product: macos
detection:
    selection_dialog:
        Image|endswith: '/osascript'
        CommandLine|contains|all:
            - 'display dialog'
            - 'hidden answer'
    selection_dscl:
        Image|endswith: '/dscl'
        CommandLine|contains: 'authonly'
    condition: 1 of selection_*
falsepositives:
    - IT automation (Jamf, Munki, custom scripts) that prompts for a password with a hidden-answer dialog
    - Admin scripts that verify a local password with dscl authonly
level: high

Known false positives:

  • Enterprise management scripts (Jamf policies, Munki pre/post-install) that prompt for a password through osascript
  • Administrators validating a local account password with dscl authonly

Tuning notes: Exclude the process ancestry of your MDM agent (Jamf, Kandji, Intune) and known IT script paths. Raise to page-level when both signals fire for the same user within 10 minutes, since that pairing is the AMOS password-capture loop. Treat any hit as a possible credential compromise: rotate the macOS account password and Keychain secrets.

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)
  • Fake AI developer-tool lure — look-alike domain and known campaign infrastructure · unlock with TL Creds
  • Gatekeeper bypass — xattr quarantine strip and execution of a binary from /tmp · unlock with TL Creds
  • AMOS credential theft — Keychain, browser store and wallet access by script or temp binary · unlock with TL Creds
  • Silent curl-to-shell installer one-liner from a non-allowlisted host (macOS ClickFix) · unlock with TL Creds
  • AMOS anti-VM / hardware fingerprinting via system_profiler or sysctl from osascript or temp binary · unlock with TL Creds
  • Web proxy traffic to AMOS / MacSync ClickFix infrastructure (C2, payload and exfiltration hosts) · unlock with TL Creds
  • AMOS ClickFix — base64-decoded curl piped to shell (sibling-process chain) · unlock with TL Creds
  • AMOS data staging — zip archive in temp directory followed by curl upload · unlock with TL Creds

Indicators of compromise

As of 2026-10-02, Threadlinqs Intelligence tracks 13 indicators of compromise for TL-2026-2916.

Sample indicators (3 of 13)

  • Domain: claude-code-cmd[.]squarespace[.]com
  • File hash: 79cd21185c51a5bfe2cfebdc51e14b258d91549fc0e4e09b6939c2a8a1c5ac19
  • Domain: download[.]active-version[.]com

Defanged, TLP:CLEAR. Full set of 13 indicators: unlock with TL Creds.

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

Technical analysis

The incident was documented with a pcap on 2026-10-02 and fits a wider 2026 pattern of AI-developer-tool malvertising against macOS users.

On 2026-10-02 Malware-Traffic-Analysis.net published a DFIR write-up of an Atomic macOS (AMOS) Stealer infection that began with a malicious advertisement impersonating Claude Code. The documented chain is: (1) a malicious ad mimicking Claude Code branding, (2) a fraudulent website spoofing Claude Code, (3) ClickFix-style instructions that tell the user to run a command in Terminal, (4) execution of that command, which installs AMOS Stealer, and (5) a request for the system password and permission grants. The page includes five screenshots, a pcap (22.8 MB), malware files (1.3 MB) and an IOC list (2.7 KB). The three archives are password-protected and were not available for review. Specific domains, IPs and hashes for this exact incident are therefore not confirmed here.

This incident matches a series of 2026 campaigns that use the same lure pattern. Bitdefender (2026-03-11) documented fake 'Claude Code' Google Ads that led to fake documentation pages. These pages delivered ClickFix commands: on macOS a base64-decoded script piped to zsh that fetched a Mach-O backdoor to /tmp/helper, with extended attributes stripped by xattr -c, and on Windows an mshta.exe-launched HTA. The campaign abused a compromised advertiser account and its macOS payload showed anti-sandbox/anti-VM checks similar to AMOS. Moonlock Lab and AdGuard (2026-02-18) described ClickFix lures hosted on malicious claude.ai artifacts promoted by Google Ads and delivering the MacSync stealer, an AMOS-family variant. A May 2026 report put that campaign at 200+ malicious ads and 35+ compromised advertiser accounts. Cato Networks (2026-08-24) reported a fake OpenAI Codex ClickFix lure on Google Sites with iframe-hosted, OS- and path-gated content, delivering a payload suspected to be AMOS and noting that similar Claude Code campaigns exist. Whether the 2026-10-02 incident belongs to the same operators is not established by the available sources.

AMOS is a malware-as-a-service macOS infostealer, first identified in April 2023 and rented on Telegram for about $1,000 per month. Public analyses (Cyble, Moonlock, K7, Sophos) describe a fake system-preferences password prompt shown through osascript, with the password validated via dscl authonly. It then steals Keychain data, browser credentials, cookies, autofill and payment data, and cryptocurrency wallet data and extensions. The data is zipped and sent to the C2 by HTTP POST. The AMOS-family technique details above come from those public analyses and have not been confirmed against this incident's samples.

Defender priorities: treat any website instruction to paste a command into Terminal as hostile, including for developer tools. Hunt for curl-to-zsh/bash pipelines launched from Terminal, osascript password dialogs, dscl authonly calls, xattr -c on files in /tmp, and zip archives created in /tmp followed by outbound POSTs. Block or alert on sponsored-search landing pages that impersonate developer tools.

MITRE ATT&CK techniques by tactic:

Collection

T1005 Data from Local System; T1560.001 Archive Collected Data: Archive via Utility

Defense Evasion

T1036 Masquerading; T1140 Deobfuscate/Decode Files or Information; T1497.001 Virtualization/Sandbox Evasion: System Checks

Credential Access

T1056.002 Input Capture: GUI Input Capture; T1539 Steal Web Session Cookie; T1555.001 Credentials from Password Stores: Keychain; T1555.003 Credentials from Password Stores: Credentials from Web Browsers

Execution

T1059.002 Command and Scripting Interpreter: AppleScript; T1059.004 Command and Scripting Interpreter: Unix Shell; T1204.004 User Execution: Malicious Copy and Paste

defense-impairment

T1553.001 Subvert Trust Controls: Gatekeeper Bypass

Resource Development

T1583.008 Acquire Infrastructure: Malvertising

Remediation

Immediate actions

  • Do not paste or run Terminal commands from websites or ads, including pages that claim to be Claude Code or other developer-tool installers
  • On any Mac where such a command was run: isolate the host, rotate the macOS account password, and rotate all credentials, session cookies and API/SSH keys stored in browsers and Keychain
  • Move cryptocurrency wallet funds to new wallets created on a clean device
  • Hunt for curl-to-zsh/bash pipelines, osascript password prompts, dscl authonly calls, xattr -c on files in /tmp, and zip archives followed by outbound POSTs

Workarounds

  • Use MDM controls to restrict unsigned or unnotarized binary execution
  • Use DNS/web filtering to block known ClickFix and malvertising infrastructure

Longer-term hardening

  • Install Claude Code only from Anthropic's official documentation and package channels, and bookmark the official URL instead of using sponsored search results
  • Deploy macOS EDR with behavioral detection for script-to-Terminal execution and Keychain access
  • Train developers on ClickFix-style lures and malvertising for AI developer tools
  • Block or monitor newly registered and look-alike domains and ad-landing pages impersonating developer tooling

Affected and fixed versions

  • Apple — macOS
    Vulnerable versions: Users who execute the pasted Terminal command (social-engineering based; no software vulnerability)

Timeline

  • Atomic macOS Stealer (AMOS) first identified as a macOS infostealer sold as malware-as-a-service on Telegram (about $1,000/month).
  • Moonlock Lab and AdGuard report ClickFix lures hosted on malicious Claude artifacts, promoted by Google Ads, delivering the MacSync (AMOS-family) stealer.
  • Bitdefender publishes research on fake 'Claude Code' Google Ads leading to ClickFix pages that deliver a Windows stealer and a macOS Mach-O backdoor.
  • Reporting on the MacSync campaign cites 200+ malicious Google Ads, 35+ compromised advertiser accounts and 15,600+ views of malicious Claude artifacts.
  • Cato Networks reports a fake OpenAI Codex ClickFix campaign on Google Sites delivering a payload suspected to be AMOS, and notes similar Claude Code campaigns.
  • Malware-Traffic-Analysis.net publishes an AMOS Stealer infection that started from a malicious ad impersonating Claude Code, with pcap, malware files and IOC list.

Sources cited

Previous and next in malware

Further reading

Drafted by the Threadlinqs research pipeline from 10 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