Microsoft: AI is now routine for North Korean hackers, mostly to fake recruiters and job candidates

Microsoft Threat Intelligence says AI use is now "high" across North Korean state hacking groups, with three groups using it in every stage of their operations — mostly to scale fake recruiters, fake job candidates and phishing lures. The attacks are still run by people, not autonomous AI. Organisations in crypto, finance, software, government and education — and anyone hiring remote IT staff — should tighten hiring checks and phishing defences now.

What happened, in plain English

On October 29, 2026, Microsoft Threat Intelligence published a report on how North Korean state-sponsored hacking groups use artificial intelligence (AI). Microsoft names these groups after weather, and "Sleet" is its label for North Korea: Jade Sleet, Coral Sleet, Jasper Sleet and five others.

Microsoft rates their overall AI adoption as high. Three groups — Jade Sleet, Coral Sleet and Jasper Sleet — have AI built into their routine work. Four more are adopting it fast.

What do they actually use it for? Mostly the same tricks as before, just faster and more convincing:

• Fake recruiters and fake companies. AI writes job offers, investment pitches and messages in fluent, well-toned English or Korean.
• Fake job candidates. Jasper Sleet uses AI to build believable online personas, résumés and interview answers, and to help with coding tests, so North Korean IT workers can get hired remotely by foreign companies and send the salaries home.
• Phishing at volume. One group, Onyx Sleet, produced more than 100 tailored PDF lure files in August and September 2026, dozens within minutes of each other.
• Malware written faster. AI coding tools help write, convert and disguise malicious code.

Think of it as a scam call centre that has hired a tireless copywriter and translator. The script is the same; there are just many more calls, and fewer spelling mistakes to give them away.

Microsoft is clear on one point: "AI-enabled operations remain human-directed and have not become fully autonomous."

Are you affected? What to do now

Microsoft says these groups mostly target cryptocurrency and financial organisations, IT and software supply chains (including code repositories), government and education. Any company that hires remote developers or contractors is a potential target for the fake-worker scheme, whatever its sector.

For most readers there is no patch to install. The work is in hiring, email and developer hygiene.

Checklist, in order:

• Tighten remote hiring checks. Microsoft recommends extra scrutiny for staff supplied by staffing companies, checking whether a candidate's social and professional profiles closely resemble other accounts, confirming phone numbers are not shared between candidates, verifying résumés and references, looking for the same persona at multiple employers, requiring cameras on in video interviews and meetings, and confirming a real, long-standing online footprint. Loop in HR and procurement, not only IT.
• Brief staff on job and investment lures. The most common hook is a recruiter, a business partner or an "interview" that leads to a document, a coding test or a meeting link. Polished language is no longer a sign that a message is genuine.
• Treat take-home coding tests and shared repositories with care. Coral Sleet has trojanised private GitHub repositories since April 2026. Developers should not run code from a recruiter or new "partner" on a work machine.
• Harden email. Microsoft recommends safe link and attachment scanning, automatic removal of malicious mail after delivery, regular phishing simulations and blocking the device code sign-in flow in Microsoft Entra ID where it is not needed.
• Harden endpoints. Turn on cloud-delivered protection, endpoint detection and response (EDR) in block mode and attack surface reduction rules.
• If you use Microsoft Defender, watch for the actor-named alerts the report lists (for example "Jade Sleet threat actor activity detected", "Sapphire Sleet actor activity") and the Defender for Cloud alert "Possible AI Tools Reconnaissance Detected". Microsoft also lists antivirus detections named ProcmacSteal tied to a Sapphire Sleet campaign.

Microsoft did not publish indicators of compromise (IOCs) such as domains or file hashes in this report.

The expert view

The useful part of this report is not the headline "attackers use AI" — that has been true for two years — but the maturity model. Microsoft sorts the eight groups into stages: operationalised (Jade, Coral and Jasper Sleet), accelerating (Diamond, Onyx, Pearl and Sapphire Sleet) and expanding (Emerald Sleet). No group is at the "emerging" stage any more. Compared with Microsoft's March 6, 2026 report, which said it had "not yet observed large-scale use of agentic AI by threat actors", the trajectory is clear: AI has moved from experiment to standard tooling for a whole state's offensive programme.

Most of the uplift is in identity deception. Translation, tone, résumé writing, interview coaching and edited photos, video and voice remove the "tells" defenders used to rely on — awkward phrasing, inconsistent stories, a face that does not match a document. Microsoft says Emerald Sleet may be using synthetic voice and video, but could not determine the operational impact. That point is still unconfirmed.

On the development side, Microsoft says AI-assisted development platforms "such as Cursor", a mainstream AI coding environment, have likely enabled four groups — Jade, Coral, Pearl and Sapphire Sleet — in code generation, conversion, debugging and obfuscation. It assesses that Coral Sleet's repository-trojanising workflow was "highly likely" engineered with an AI development platform, and that Sapphire Sleet's malware in an August 2026 compromise of Rust software was markedly better than its earlier work. The pattern matters: the tools are ordinary developer tools, not underground "evil AI". Faster retooling means shorter shelf-lives for signature-based detections.

The most notable new item is agentic. Microsoft says that since late October 2026 it has seen Onyx Sleet and Pearl Sleet use ARTEX, which it describes as an open-source agentic AI penetration-testing framework. ARTEX already appeared in other reporting in October 2026: Microsoft linked it to the China-aligned group Storm-0940, and CrowdStrike has reported its use against financial institutions in South Korea. North Korean groups picking up the same publicly released offensive agent shows how quickly such tools spread between unrelated actors once published. It also shows the limit of "made private later" as a control — the repository's developer reportedly restricted it on October 8, 2026, but copies were already out.

What is not new: initial access still relies on social engineering, malware delivery and abuse of legitimate tools. The motives — revenue (largely crypto theft and IT-worker salaries), intelligence collection and occasional disruption — are unchanged. Defences that already work against these groups (phishing-resistant authentication, controlled code execution, identity verification in hiring) keep working; AI raises the volume and polish, so gaps get found faster.

Still unknown: the report does not name the AI models or providers behind the text and media generation, does not say whether any accounts were disrupted, and gives no victim counts. The ARTEX observation is very recent, so its scale and success rate are not yet clear. All attributions in this article are Microsoft's.

Official sources

• Microsoft Threat Intelligence — "Tracking AI adoption in North Korean threat actor tradecraft" (October 29, 2026)
• Microsoft Threat Intelligence — "AI as tradecraft: How threat actors operationalize AI" (March 6, 2026)
• Coverage: TokenPost — Microsoft Identifies AI Use Across Eight North Korean Hacker Groups
• Coverage: Fortune — "Black October" AI cyberattacks overview (October 29, 2026)