Microsoft rates AI adoption by eight North Korean hacking groups, three fully operational

What Microsoft Threat Intelligence published

Microsoft Threat Intelligence published a report on 29 October 2026, Tracking AI adoption in North Korean threat actor tradecraft, that rates eight state-linked groups from the Democratic People's Republic of Korea on how far they have built generative and agentic AI into their work. Microsoft calls overall adoption high and describes AI as a force multiplier that speeds up workflows the groups already ran, not a change in their goals: raising revenue for the sanctioned regime, collecting intelligence and, in some cases, disruptive attacks.

The report uses four stages. Jade Sleet, Coral Sleet and Jasper Sleet are at the top, operationalized stage, with AI in every routine workflow. Diamond Sleet, Onyx Sleet, Pearl Sleet and Sapphire Sleet are accelerating. Emerald Sleet is expanding, with AI in some workflows only. No group was placed in the lowest, emerging stage.

Microsoft lists the usual targets as cryptocurrency and financial firms, IT companies and software supply chains, government bodies and education organisations. AI also appears alongside shifts in targeting: since August 2026, Diamond Sleet has returned to the oil and gas sector for the first time since 2023 and built fake communications technology companies, complete with websites and personas, to approach telecom firms as supposed business partners. Microsoft says AI is used across research on organisations and employees, crypto wallets and exchanges, and vulnerabilities, APIs and cloud services, producing a visible change in the speed and sophistication of activity against specific technologies.

The key caveat concerns autonomy. Microsoft writes that the AI-enabled operations remain human-directed and have not become fully autonomous: operators set intent and objectives, while AI handles research, content, code and repetitive tasks. The data describes a large productivity gain for existing teams rather than self-directed attack agents.

Phishing lures, fake personas and coding tools

The clearest numbers come from Onyx Sleet. Microsoft identified more than 100 PDF lure files attributed to the group, created in August and September 2026 with target-specific details; metadata showed dozens made within minutes, which Microsoft calls highly likely to be AI output. The lures supported high-volume phishing against South Korean users. Between 8 and 12 June 2026, Jade Sleet posed as recruiters for a spoofed AI-powered blockchain intelligence company and ran similar job-themed campaigns through two more front companies in the same week.

Jasper Sleet, linked to fraudulent remote employment, uses generative and agentic AI to build resumes, personas and interview answers and to keep several fake worker identities consistent. Microsoft also reports possible synthetic voice and video generation by Emerald Sleet, without a determined impact.

On development, Microsoft says AI coding platforms such as Cursor likely helped Jade Sleet, Coral Sleet, Pearl Sleet and Sapphire Sleet write, debug and obfuscate code. Sapphire Sleet likely used AI in an August 2026 compromise aimed at Rust developers, and since April 2026 Coral Sleet has trojanized private GitHub repositories with a workflow Microsoft judges was highly likely built with AI. Since late October 2026, Onyx Sleet and Pearl Sleet have used ARTEX, the open-source agentic penetration testing framework Microsoft linked to China-aligned Storm-0940 on 21 October.

What defenders are told to do, and what is not stated

Because several groups target hiring, Microsoft's most practical advice is for recruiters: scrutinize workers supplied by staffing companies, check that profiles and phone numbers are unique, call references by phone or video, look for the same persona employed at several companies, require candidates on camera in several sessions and confirm a real digital footprint with a non-VoIP number and residential address. It also recommends phishing simulations, zero-hour auto purge, Safe Links, blocking device code flow, endpoint detection in block mode and attack surface reduction rules, and points to its own Security Dashboard for AI and Agent 365, products it sells.

Microsoft's guidance is mostly about older techniques that AI now makes faster and cheaper to run at volume. The company warns that rapidly regenerated lures make blocking and disruption harder, and that AI-assisted development lets the groups retool quickly when defenders detect a technique. Controls that rely on spotting clumsy language, slow attackers or reused files are therefore likely to weaken.

The report does not give victim counts, stolen amounts or the generative AI model providers the groups used, apart from coding tools such as Cursor. It does not say whether AI vendors suspended accounts. The stage ratings are Microsoft's own assessment. The practical reading for security teams, after a month that also brought a backdoored Claude Code extension, is that phishing and fake candidates will keep getting faster and more polished.

Questions people ask

Which North Korean groups does Microsoft say use AI the most?

Microsoft places Jade Sleet, Coral Sleet and Jasper Sleet at the operationalized stage, meaning AI is part of all their routine workflows. Diamond Sleet, Onyx Sleet, Pearl Sleet and Sapphire Sleet are at the accelerating stage, and Emerald Sleet at the expanding stage.

Are North Korean hackers running fully autonomous AI attacks?

No, according to Microsoft. Its 29 October 2026 report says the AI-enabled operations remain human-directed and have not become fully autonomous, although two groups now use the agentic penetration testing framework ARTEX.

How can companies spot AI-assisted fake job applicants?

Microsoft advises checking that profiles and phone numbers are unique, verifying names, addresses and dates, calling references by phone or video, requiring candidates on camera in several sessions and confirming a real digital footprint with a non-VoIP number and residential address.

Sources

1. Tracking AI adoption in North Korean threat actor tradecraft (Microsoft Threat Intelligence, 29 Oct 2026)
2. Storm-0940 uses ARTEX agentic AI in ConnectWise supply chain compromise (Microsoft Threat Intelligence, 21 Oct 2026)