Skip to content

Response, New York. Noah Park: Check Point Patches Actively Exploited Management Server Path Traversal. Detection, Dallas. Ava Okello: Report Finds Nearly Half of Deployed Detections Need Attention Before SOCs Can Trust Them.

SOCtember

Always First. Fast SOC News.

Microsoft Threat Intelligence header image for its report on AI adoption by North Korean threat actors, showing a data center aisle with the Sleet actor icon.
Image: Microsoft Threat Intelligence (fair use). Header art from Microsoft's Oct. 29 report on how North Korean threat actors use AI.

Threat Intel

Microsoft Says North Korean Hackers Have Built AI Into Their Operations, With Onyx Sleet Producing More Than 100 Tailored PDF Lures and Two Groups Now Using the ARTEX Agentic Pentest Framework

Microsoft Threat Intelligence rated AI adoption high across the regime's actors and put Jade Sleet, Coral Sleet and Jasper Sleet at the operationalized stage. It said the operations remain human-directed, and it published Defender detections and hiring checks aimed at fake IT worker personas.

Priya Shah, Threat IntelSingapore5 min read

SINGAPORE - North Korean state hackers have folded generative and agentic AI into almost every stage of their operations, from researching targets to writing malware, and three of the regime's groups now use it as a routine part of their work, Microsoft Threat Intelligence said Thursday.

In a report published on Oct. 29, Microsoft rated overall AI adoption across North Korean threat actors as high. It said Jade Sleet launched campaigns through three front companies in a single week, Onyx Sleet generated dozens of tailored lures in minutes and Coral Sleet rapidly built a capability to trojanize private source code repositories. It also said Onyx Sleet and Pearl Sleet have begun using ARTEX, the open-source agentic AI penetration testing framework, in their operations.

Microsoft was careful about the limits of what it saw. It "assesses that AI-enabled operations remain human-directed and have not become fully autonomous," the report said, and it described AI as a force multiplier for established workflows rather than a change in the regime's goals of raising revenue, collecting intelligence and, in some cases, disruption.

Four stages, no beginners

Microsoft diagram placing North Korean threat actors on four stages of AI adoption: emerging, expanding, accelerating and operationalized.
Image: Microsoft Threat Intelligence (fair use).Microsoft's map of North Korean actors by stage of AI adoption.

Microsoft sorted the actors it tracks into four stages. Jade Sleet, Coral Sleet and Jasper Sleet are at the operationalized stage, with AI embedded across their entire operations. Diamond Sleet, Onyx Sleet, Pearl Sleet and Sapphire Sleet are accelerating. Emerald Sleet is expanding, using AI in some workflows without the same level of adoption and impact. Microsoft said it has not assessed any North Korean actor as being at the emerging, or experimental, stage.

What the AI is doing

Between June 8 and June 12, 2026, Jade Sleet posed as recruiters for a spoofed AI-powered blockchain intelligence company, using software engineer job offers to reach people at cybersecurity, AI, blockchain and talent recruitment companies, and ran similar job-themed outreach through two more front companies in the same window. Microsoft said the pace and breadth of that activity show the group using generative AI to build industry knowledge and scale outreach.

Onyx Sleet produced more than 100 PDF lure files with target-specific details in August and September 2026. File metadata showed dozens created in rapid succession within minutes, which Microsoft called "highly likely to be the output of an AI platform." The lures support high-volume phishing against South Korean users, and Microsoft warned that rapid regeneration of lures could make detection and disruption harder.

Jasper Sleet, the group behind fraudulent remote IT worker schemes, uses generative AI platforms to build fake digital personas, with what Microsoft suspects are generative and agentic AI-assisted resume workflows, and its tooling helps keep communication consistent across multiple fake worker identities. Microsoft also identified possible AI-generated voice and video from Emerald Sleet.

On the development side, Microsoft said AI coding platforms such as Cursor have likely helped Jade Sleet, Coral Sleet, Pearl Sleet and Sapphire Sleet generate, debug and refactor code, write persistence and obfuscation, and port tools across languages. Sapphire Sleet likely used AI in the August 2026 Rust software compromise, where its malware used dynamic .NET compilation during post-exploitation, most likely for defense evasion. Since April 2026, Coral Sleet has run a workflow to trojanize private GitHub repositories that Microsoft said was "highly likely engineered using an AI-enabled development platform."

Diamond Sleet has returned to targeting the oil and gas sector since August 2026, for the first time since 2023, and has set up fake communications technology companies, with websites and personas, to approach telecommunications providers as would-be business partners.

What defenders get

Microsoft Defender detection table from the report, listing actor-specific alerts for seven Sleet groups and reconnaissance and sign-in detections.
Image: Microsoft Threat Intelligence (fair use).The report's Defender coverage table.

The report lists Defender for Endpoint alerts for seven of the groups; Defender for Cloud alerts named "Possible AI Tools Reconnaissance Detected" and "Possible Secret Reconnaissance Detected"; a Defender XDR alert for sign-in activity by a suspected North Korean entity tied to Jasper Sleet; Entra ID Protection atypical and impossible travel alerts; and antivirus detections for the backdoors seen in the Sapphire Sleet Rust compromise, named ProcmacSteal across Linux, PowerShell, macOS and Cargo variants.

Its mitigation list leans on controls that SOC and identity teams already own: block device code flow wherever possible, turn on zero-hour auto purge, Safe Links and Safe Attachments, run EDR in block mode, and enable attack surface reduction rules that block executable content from email, obfuscated scripts and low-prevalence executables.

Microsoft's guidance for vetting potential employees, including checks on shared phone numbers, on-camera interviews and a real digital footprint.
Image: Microsoft Threat Intelligence (fair use).The report's hiring checks aimed at fake worker personas.

Because so much of this activity runs through hiring, Microsoft also gave HR-facing checks: apply greater scrutiny to staffing company employees, confirm that a candidate's contact phone number is not shared with other accounts, contact references by phone or video, search for the same persona employed at multiple companies, and make sure candidates appear on camera in multiple sessions. A candidate who reports video or microphone problems that prevent participation "should be considered a red flag," the report said.

What the SOC should take from it

For threat intelligence and detection teams, the practical point is volume and speed. Lures that can be regenerated in minutes make static indicators age quickly, so the report's own artifacts point toward behavior: job-themed contact that moves to code assessments or repositories, private repositories that change unexpectedly, device code sign-ins that the business does not use, and new hires whose identity signals do not hold together. The report also arrives eight days after Microsoft's Oct. 21 account of China-aligned Storm-0940 using ARTEX to accelerate the ConnectWise supply chain compromise, so hunts written for ARTEX activity now cover two North Korean groups as well.

Sources:


Priya Shah covers threat intelligence, intrusion analysis, and adversary tradecraft for SOCtember from Singapore.

Related stories

Threat Intel desk