Who they are and who they target
Assessment date: November 22, 2026. Everything below reflects public reporting and government attribution available on that date. Where this profile says confirmed, a government attribution, a court document or a named advisory exists. Where it says assessed, the private sector or the wider intelligence community has reached a judgment with varying confidence, and you should treat the detail as a working hypothesis rather than fact.
Lazarus Group is not one team. It is an umbrella label for a set of North Korean state-sponsored intrusion clusters that share infrastructure, tooling and handlers. The US government calls the same activity Hidden Cobra, and the Treasury’s Office of Foreign Assets Control sanctioned the Lazarus Group in September 2019 alongside two related clusters, Bluenoroff and Andariel. That designation, plus the FBI’s public attribution statements, is the strongest public basis for tying the activity to the North Korean state, specifically the Reconnaissance General Bureau, the country’s foreign intelligence service.
The umbrella is the problem. Commercial vendors split the same activity into Diamond Sleet, Sapphire Sleet and Onyx Sleet (Microsoft), TraderTraitor (CISA and the FBI), and APT38, which MITRE ATT&CK tracks as a separate group (G0082) from Lazarus (G0032) even though both map to North Korean operations. A report about one name does not automatically describe the others, and cluster boundaries move. Attribution caveat: a specific intrusion is confirmed only when a government or a court says so; the mapping of a given malware sample to a given sub-cluster is nearly always an assessment.
Confirmed by name: the 2014 Sony Pictures destructive attack (FBI, December 2014), the 2016 Bangladesh Bank SWIFT theft of about $81 million (US Department of Justice indictment of a North Korean programmer, 2018), the 2017 WannaCry outbreak (attributed by the US and UK governments), the 2022 Ronin bridge theft (FBI), and the February 2025 Bybit theft of roughly $1.5 billion in cryptocurrency (FBI, attributed to TraderTraitor). This profile cites only cases backed by a government attribution or a named advisory, so the most recent case it cites is from February 2025.
Who they go after follows the money and the mission. Banks and payment operators. Crypto exchanges, DeFi bridges, wallet and custody providers. Defense, aerospace and engineering firms, usually through fake job offers. Healthcare, when the goal is disruption rather than theft. And developers: package maintainers, blockchain engineers, and anyone willing to open a “coding challenge” from a recruiter.
Known tooling and tradecraft
Two motivations shape the toolset: espionage and sabotage for the state, and revenue for a sanctions-constrained economy. That produces two styles of intrusion that share infrastructure.
- Wipers and destructive malware. The Sony attack used wipers that destroyed data and boot records on a large scale. Confirmed by the FBI.
- Ransomware as cover. WannaCry (2017) spread through the EternalBlue SMB exploit and hit hospitals and manufacturers worldwide; the US and UK attributed it to North Korea. Maui ransomware, used against healthcare providers, was attributed to DPRK actors in a 2022 CISA and FBI advisory. Assessed.
- Custom backdoors and tunnels. CISA and FBI malware advisories have published indicators for families including Fallchill, BLINDINGCAN, HOPLIGHT and the ELECTRICFISH traffic tunnel. These are useful for indicator-driven hunting, but they age quickly.
- AppleJeus. Fake crypto-trading companies distribute trojanized installers for Windows and macOS (CISA advisory AA21-048A, February 2021). The lure is a convincing product page, not a phishing email.
- TraderTraitor. Social engineering of blockchain and exchange employees, often through fake job offers, followed by malicious trading or portfolio apps (CISA and FBI advisory AA22-108A, April 2022).
- Contagious Interview. A recruitment-themed campaign documented by Unit 42 since late 2023, delivering the BeaverTail JavaScript infostealer and the InvisibleFerret Python backdoor through fake interview and coding-test pages.
- Supply chain abuse. Microsoft reported in November 2023 that its Diamond Sleet cluster compromised a legitimate software vendor to distribute trojanized installers, and used signed binaries and DLL side-loading to stay resident.
Common threads worth detecting on: valid code-signing certificates, sometimes stolen and sometimes fraudulently obtained; DLL side-loading into trusted applications; living-off-the-land binaries; archives and disk images as delivery containers; and command-and-control hidden behind legitimate cloud services or newly registered domains.
Techniques and detection
Map detections to ATT&CK, then hunt sequences rather than single events. The techniques below are where Lazarus-style operations leave the most usable evidence.
- T1195.002, Compromise Software Supply Chain. Hunt for signed applications loading unsigned DLLs from user-writable paths, installers whose publisher or hash is not on your allowlist, and update traffic to domains that do not match the vendor’s registered infrastructure.
- T1566.003 and T1566.002, spearphishing through a third-party service or a link. Recruiter or job-board conversations followed within hours by archive downloads, links to domains registered recently, and messaging traffic to file-sharing or code-hosting services.
- T1204.002, Malicious File. Office or PDF readers spawning scripting hosts, archives containing LNK files or executables, and documents that demand content be enabled before anything visible happens.
- T1059.001 and T1059.006, PowerShell and Python. Encoded commands, download cradles, amended execution policy, and python.exe or node.exe making outbound HTTPS from user directories.
- T1543.003, Windows Service. Service creation events pointing at binaries in ProgramData, AppData or Temp, especially shortly after an installer runs. This is the BLINDINGCAN-style persistence pattern.
- T1555.003, Credentials from Web Browsers. Non-browser processes reading browser Login Data, Cookies or Local State files. That is BeaverTail’s core behavior and it is rare in a clean estate.
- T1105 and T1071.001, tool transfer over web protocols. Repeated small HTTPS POSTs to a single host from developer or finance endpoints, and connection timing with regular jitter.
Proxy execution through signed binaries is cheap for the attacker and rare in most estates. Run this with a thirty-day lookback, then tune:
DeviceProcessEvents
| where FileName in~ ("rundll32.exe","regsvr32.exe")
| where ProcessCommandLine has_any (@"\AppData\", @"\ProgramData\", @"\Temp\")
| where InitiatingProcessFileName in~ ("winword.exe","excel.exe","chrome.exe","msedge.exe","python.exe")
| project Timestamp, DeviceName, AccountName, FileName, ProcessCommandLine
Add two graph-style hunts that do not depend on signatures: Sysmon image-load events where a signed executable loads a DLL from a non-system path, and any process that opens a browser credential store. Both are low volume, and both are expensive for the actor to avoid while still achieving the objective.
Mitigation priorities
Prioritize by what the actor does with access. In confirmed cases the objective is theft or destruction, not quiet persistence for its own sake.
- Make one compromised laptop insufficient to move money. Hardware-backed keys, multi-party approval for transactions, out-of-band verification of payout addresses, and a hard rule that signing keys never live on a workstation that reads email or chat. This matters most because crypto theft is the confirmed, repeated objective.
- Treat recruitment as an attack surface. Verify candidate identity beyond a video call, run coding tests in a disposable VM with no corporate credentials, and block execution of anything a candidate sends. Teach engineers that an unsolicited recruiter with a perfect role is a phishing pretext until proven otherwise.
- Constrain software supply chains. Allowlist applications and update channels, verify installers against vendor-published hashes, pin dependencies, and alert on new browser extensions and IDE plugins.
- Break credential and browser theft. Phishing-resistant MFA, credential isolation for administrators, endpoint detection that flags non-browser access to browser stores, and no stored passwords in browsers on finance or engineering machines.
- Application control and script logging. WDAC or AppLocker rules covering user-writable paths, PowerShell script block logging, AMSI, and an agreed policy for Python and Node execution on developer endpoints.
- Egress control and DNS hygiene. Restrict outbound traffic from finance and engineering segments to known destinations, alert on newly registered domains, and inspect TLS metadata where lawful. Patch internet-facing edge devices on a schedule you can prove.
- Rehearse the response. Decide in advance who calls the FBI, who calls counsel, and who can freeze wallets. Confirmed DPRK cases routinely turn on attribution and chain of custody, so preserve volatile data before remediation.
What to watch for next
Assessed, not confirmed. These are expectations to test against your telemetry, not reporting.
- More developer-tooling supply chain. Package registries, IDE plugins and CI credentials are the highest-leverage targets still standing.
- Better AI-assisted lures. Recruitment outreach that is localized, role-specific and grammatically clean. Detection has to rest on behavior, not on clumsy phrasing.
- macOS and cloud identity. AppleJeus demonstrated the macOS intent. Expect more pressure on SSO sessions, where a stolen token removes the need for malware entirely.
- Ransomware as deniability. WannaCry and Maui show the pattern of destructive operations dressed as crime. A ransom note is not proof of a criminal actor.
- IT-worker infiltration. DPRK nationals posing as remote contractors is a related, growing problem flagged by Treasury guidance and vendor research. It needs HR verification and identity controls, not just endpoint detection.
Two rules for the coming year. First, stop organizing detection around group names and organize it around behaviors that survive re-clustering: side-loading, credential-store access, signed-binary proxy execution, and archive-delivered loaders. Second, read the primary advisories rather than summaries. CISA AA21-048A, AA22-108A and the FBI attribution statements are short, specific and free.
For the group’s origins and how the naming evolved, our Lazarus Group backgrounder is the companion piece to this assessment.