The Pulse
Microsoft Finds North Korean Groups Using ARTEX AI Agents
Microsoft says North Korean groups Onyx Sleet and Pearl Sleet used the open-source ARTEX agentic penetration-testing framework in operations. Its October 29 assessment also describes AI-assisted phishing, fake identities and malware develop
AI.info Team · Written with AI · fact-checked ·
“Current evidence indicates that these actors primarily use AI to enhance existing tradecraft and operator productivity rather than to enable autonomous operations.”
Microsoft Threat Intelligence
Microsoft says two North Korean threat groups, Onyx Sleet and Pearl Sleet, have used an open-source agentic AI penetration-testing framework in operations. The company’s October 29 assessment names the tool as ARTEX and places its reported use within a broader campaign of AI-assisted reconnaissance, identity deception and software development.
The finding marks a specific step beyond using generative AI to write lures or research targets: Microsoft says the groups have brought an agentic framework into operational activity. But its report does not describe ARTEX independently choosing targets or carrying out attacks without people. Microsoft assesses that AI-enabled operations remain human-directed.
Onyx Sleet and Pearl Sleet adopt ARTEX
Microsoft Threat Intelligence says it identified Onyx Sleet and Pearl Sleet using ARTEX “since late October 2026.” The company describes ARTEX as an open-source agentic AI penetration-testing framework, but does not detail which tasks the two groups performed with it, name specific victims of those operations or say that ARTEX carried out a complete intrusion autonomously.
That limit matters. The report documents adoption of an agentic tool in operations, not a fully autonomous cyberattack. Microsoft says human operators likely continue to direct intent and objectives, even as AI assists with repetitive work, information summaries and decisions after a compromise.
AI helps scale phishing and false identities
Microsoft’s examples also cover less autonomous uses. Between June 8 and June 12, 2026, the company observed Jade Sleet using job-themed approaches against people in cybersecurity, AI, blockchain and recruitment organizations. The group posed as recruiters for a fake AI-powered blockchain intelligence company, then used two additional front companies with similar infrastructure and industry-specific knowledge.
Microsoft attributes the pace and tailoring of that activity to generative AI-assisted workflows. It also says it found more than 100 target-specific PDF lure files attributed to Onyx Sleet, created in August and September. Metadata showed dozens of files appearing within minutes; Microsoft judged it highly likely that an AI platform produced them.
The report describes AI assistance with fake resumes, biographies, interview answers, translations and profile materials, as well as support for managing multiple worker identities. Microsoft says Jasper Sleet uses AI tooling to maintain communication across fraudulent worker personas. These methods support social engineering and attempts to obtain legitimate remote employment, rather than replacing human operators with independent agents.
Three groups reach Microsoft’s top adoption tier
Microsoft sorts eight named North Korean threat groups into four stages of AI adoption: emerging, expanding, accelerating and operationalized. It puts Jade Sleet, Coral Sleet and Jasper Sleet in the operationalized tier, describing AI as embedded across their operations. Diamond Sleet, Onyx Sleet, Pearl Sleet and Sapphire Sleet are in the accelerating tier; Emerald Sleet is expanding its use.
The assessment also links AI-assisted development to malware and supply-chain activity. Microsoft says Sapphire Sleet likely incorporated AI during an August 2026 compromise involving Rust software, and that Coral Sleet had, since April, built a workflow for trojanizing private GitHub repositories. The company says AI development tools can help generate or refactor code, troubleshoot errors and adapt tools—capabilities that may help attackers change tactics when defenders block them.
The report draws a line at autonomy
Microsoft’s central claim is that AI increases the speed and scale of established North Korean operations, not that it has changed their aims. The company says those aims include generating revenue, collecting strategic intelligence and, in some cases, pursuing disruptive or destructive objectives. It identifies cryptocurrency and financial firms, technology companies, software supply chains, government bodies and education organizations among common targets.
ARTEX’s reported use is the clearest agentic-AI finding in the post, but Microsoft provides little operational detail about the framework’s role in those cases. Its broader evidence describes AI accelerating reconnaissance, phishing, persona management and technical work, while people remain in control of the objectives. In Microsoft’s own classification, three of the eight groups it assessed have embedded AI across their operations.