Tracking AI adoption in North Korean threat actor tradecraft

What it is

A Microsoft Threat Intelligence report published 29 October 2026. It assesses overall North Korean AI adoption as high and says AI mostly speeds up existing social engineering, malware and reconnaissance work. Microsoft judges operations to be human-directed and not autonomous, and says AI has not changed the actors' goals: revenue, strategic intelligence, and some disruptive activity. Targets remain crypto and finance, IT firms and software supply chains, government, and education.

How it works

Observed use, by activity:

• Reconnaissance and personas: From 8 to 12 June 2026, Jade Sleet posed as recruiters for a spoofed AI-powered blockchain intelligence company, then ran the same job-themed approach through two more front companies in that window. Jasper Sleet uses suspected generative and agentic resume workflows and repeatable tooling to manage multiple fake worker personas. Since August 2026, Diamond Sleet has returned to oil and gas for the first time since 2023, widened to communications and internet policy organizations, and built fake telecom technology companies to target telecoms through "business partnerships".
• Lures: Over 100 target-specific Onyx Sleet PDFs from August and September 2026, with metadata showing dozens created within minutes. They support high-volume phishing of South Korean users. Microsoft says rapid regeneration could make disruption and detection harder. Emerald Sleet shows possible AI voice and video generation, with impact undetermined.
• Tooling: Cursor-style AI development platforms likely help Jade Sleet, Coral Sleet, Pearl Sleet and Sapphire Sleet write, debug, convert and obfuscate code and retool after defenders block them. Sapphire Sleet likely used AI in the August 2026 Rust software compromise, and its malware used dynamic .NET compilation during post-exploitation, most likely for evasion. Since April 2026 Coral Sleet has run a workflow that trojanizes private GitHub repositories and feeds them into its malware chain.
• Agentic tooling: Since late October 2026, Onyx Sleet and Pearl Sleet have used the open-source agentic pentest framework ARTEX in operations.

Preconditions / Affected

Exposure is highest for crypto, finance, software and AI firms, telecoms, defense, oil and gas, and organizations that hire remote developers or use staffing firms. South Korean users face the Onyx Sleet phishing.

What to do

The guidance is general hardening, mostly Microsoft-product specific:

• Block device code flow wherever possible, and where it is needed restrict it through Entra Conditional Access.
• Turn on Safe Links, Safe Attachments and Zero-hour auto purge in Defender for Office 365, and run phishing simulations.
• Enable cloud-delivered protection, EDR in block mode, automated investigation and remediation, automatic attack disruption, network protection and attack surface reduction rules.
• Put web-facing systems behind a VPN, WAF, reverse proxy or DMZ, and automate patching.
• For hiring, check for duplicate social profiles or phone numbers, reject VOIP numbers, require camera on across several video sessions, treat claimed camera or microphone faults as a red flag, call references by phone or video, and check whether the same persona works at several companies. Give staffing-company hires extra scrutiny.
• Microsoft also promotes its Security Dashboard for AI and Agent 365 for governing AI agents.

Caveats

The page gives no indicators of compromise, hashes, domains or hunting queries in the fetched text. The attack surface reduction rule list and the detection list are empty there. Much attribution of AI use is inferred ("likely", "highly likely") from metadata, pace and tooling quality, not from observed prompts. The Sapphire Sleet and Coral Sleet AI links are assessments, not confirmed facts.