Microsoft Threat Intelligence said on October 29 that three North Korean state-backed hacking groups, Jade Sleet, Coral Sleet and Jasper Sleet, "have operationalized AI across their entire operations", the top stage of a four-stage scale it applied to eight groups. Four more groups sit one stage below, and Microsoft placed none of the eight at its lowest, experimental stage.
For software companies, hiring is one of the widest doors. By SaasBuzz's count of the groups Microsoft names, four of the eight use fake recruiters, job interviews or fake job applicants, and four use AI coding platforms "such as Cursor" to write and rework malware. Microsoft also assesses that the AI-enabled operations "remain human-directed and have not become fully autonomous."
What Microsoft found
Fake recruiters at speed. Between June 8 and June 12, 2026, Jade Sleet posed as recruiters for "a spoofed AI-powered blockchain intelligence company," offering software engineering roles to people at cybersecurity, AI, blockchain and talent recruitment organizations. In the same window it ran similar outreach through two more front companies.
Lures by the dozen. Microsoft found more than 100 PDF lures attributed to Onyx Sleet, created in August and September with target-specific details. File metadata showed dozens made "in rapid succession within minutes," which Microsoft called "highly likely to be the output of an AI platform." The campaign targeted South Korean users.
Poisoned code. Since April 2026, Coral Sleet has run a workflow that trojanizes private GitHub repositories, which Microsoft said was "highly likely engineered using an AI-enabled development platform." Sapphire Sleet likely used AI in an August 2026 compromise that targeted Rust developers.
Fake employees. Jasper Sleet uses generative AI to build fake personas and resumes and to keep several fraudulent remote workers consistent.
Agents. Since late October, Onyx Sleet and Pearl Sleet have used ARTEX, an open-source agentic penetration testing framework. CrowdStrike tied the same framework to a China-nexus group's campaign against more than 200 organizations.
Microsoft's threat intelligence account summarized the findings on X when the report went up, putting the adoption down to the regime's existing goals rather than new ones.
Post on X by @MsftSecIntel
AI adoption by North Korean threat actors largely supports existing goals and targeting priorities, including generating revenue for the regime, collecting strategic intelligence, and, in some cases, pursuing politically motivated disruptive or destructive objectives, rather than representing a shift in actor motivations. https://msft.it/6010a0LAm Recent activity by these threat actors illustrates AI-enabled acceleration and scaling: - Jade Sleet launched campaigns through three front companies in a single week - Onyx Sleet generated dozens of tailored lures in minutes - Coral Sleet rapidly built a new capability to trojanize private source code repositories - Diamond Sleet used AI to enhance tailored targeting workflows, including persona development and front company creation - Pearl Sleet and Onyx Sleet have been leveraging the open-source agentic AI penetration testing framework ARTEX to conduct operations Our latest blog breaks down how North Korean threat actors are utilizing AI across their operations, and provides mitigation guidance and Microsoft Defender detections to help organizations defend against these threats.
Where each group stands
Group | Stage | AI use named |
|---|---|---|
Jade Sleet | Top | Fake recruiters, front firms, coding tools |
Coral Sleet | Top | Job lures, trojanized repos, coding tools |
Jasper Sleet | Top | Fake remote workers |
Sapphire Sleet | Second | Job lures, Rust attack, coding tools |
Pearl Sleet | Second | Coding tools, ARTEX |
Onyx Sleet | Second | PDF lures, ARTEX |
Diamond Sleet | Second | Fake comms tech firms |
Emerald Sleet | Third | Possible fake voice, video |
Source: SaasBuzz tally of Microsoft Threat Intelligence, October 29, 2026. Stages, top to bottom: operationalized, accelerating, expanding, emerging. Microsoft placed no group at the fourth.
What it means for software companies
Microsoft lists "information technology companies and software supply chains" among the groups' usual targets, and the methods it describes run through places software teams trust: a recruiter's message, a coding test, a new hire's laptop and a private repository. Among the uses it lists is "job-themed payload-delivery content and malicious repository lures," so a code repository sent by an unfamiliar recruiter deserves the suspicion of any unexpected attachment. On the other side of hiring, the same list includes help with "coding assessments" and "the management of multiple worker personas," the fake applicant's kit.
Speed is the change. Microsoft says AI platforms "streamline retooling," letting the groups "rapidly pivot in reaction to defensive measures," and that tools such as Cursor helped them generate, debug and refactor code and move it "across languages, libraries, and environments." Microsoft does not say Cursor itself was flawed or breached; it names Cursor as an example of the kind of tool the groups used. A blocked sample gets replaced faster, which favors controls on behavior and access over lists of known bad files.
Clumsy writing fades as a warning sign too. Microsoft says the groups use AI to translate and "adjust tone" for "more natural communications," and to keep recruiter, developer and investor identities going "across interactions and platforms." SaasBuzz's earlier pieces on attacks through AI coding tools and on the Binance theft tied to North Korea cover the same pressure from other sides. Microsoft's post does not mention Binance.
Microsoft's checklist, minus the product names
Microsoft gives 27 steps under four headings, many framed around its own Defender and Entra products. These apply to any software company:
Give workers supplied by staffing companies extra scrutiny, which Microsoft calls "the easiest avenue for North Korean workers to infiltrate target companies."
Check that a candidate's social and professional accounts are not near-copies of other people's, and that their phone number is not shared with other accounts.
Match names, addresses and dates across resumes and background checks, and call references by phone or video rather than email alone.
Search whether the same persona is employed at several companies.
See the candidate on camera in several video calls, and treat reported camera or microphone trouble as a red flag.
Confirm a real phone number rather than VoIP, a residential address and social media accounts.
Teach staff to ignore unsolicited attachments sent through messaging apps and social networks, and run phishing simulations.
Block the device code sign-in flow wherever possible.
Keep web-facing systems behind a VPN, or behind a web application firewall or reverse proxy where they must be public, and automate patching.
Run endpoint detection in blocking mode, and block executable content from email and obfuscated scripts.
