Microsoft: North Korean Groups Routinely Use AI in Cyber Operations

AI used by North Korean actors has, according to a new Microsoft Threat Intelligence assessment, become a routine part of reconnaissance, social engineering, tool development and the scaling of cyber operations. On October 29, Microsoft said it rates adoption as high among the groups it tracks, but the campaigns described remain human-directed and do not involve fully autonomous attacks.

The company specifically names Jade Sleet, Coral Sleet and Jasper Sleet, which it says have routinely incorporated artificial intelligence into their workflows. The published assessment is based primarily on Microsoft telemetry and analytical conclusions; individual attributions and the extent of AI use have therefore not been independently verified forensically in every incident.

AI used by North Korean actors in lures and tools

One specific example involves the Onyx Sleet group. Microsoft attributes more than 100 PDF lures created in August and September 2026 to the group. According to the company, document metadata suggests the bulk creation of dozens of files within several minutes. This approach could make it easier for attackers to prepare larger numbers of document variations intended for phishing or other forms of social engineering.

Microsoft also says that Onyx Sleet and Pearl Sleet use ARTEX, an open-source agentic framework intended for penetration testing. Based on the published information, use of this tool alone does not confirm autonomous execution of attacks. Microsoft explicitly distinguishes between the expanded deployment of AI to support operators’ work and attacks that would proceed without human direction.

In the case of the Emerald Sleet group, the company is tracking possible use of synthetic voice and video. However, it has not yet determined the full operational impact of such methods. The findings therefore do not confirm that synthetic content was a decisive factor in specific operations.

What Microsoft says is changing

The assessment does not claim that AI itself created a new offensive capability. Microsoft says it primarily accelerates existing processes: reconnaissance preparation, the creation of more convincing text and lures, code modifications and operational tasks. The volume of materials tailored to specific targets may therefore increase during attacks.

The risk particularly affects organizations facing recruitment scams, spear-phishing or supply-chain attacks. Variable PDF documents, false identities and modified malicious code can make it harder to recognize recurring campaign patterns. However, Microsoft’s published assessment does not provide specific indicators of compromise, detection rules or fixes tied to the activities described.

What has not yet been confirmed

Although Microsoft describes a high level of AI adoption, a more detailed picture will require independent technical confirmation of individual campaigns and attributions. It will also be necessary to monitor whether other security providers or government agencies confirm the use of the ARTEX framework and the extent of AI-generated lures.

Earlier findings from the Google Threat Intelligence Group had already pointed to the misuse of generative AI by actors linked to North Korea for reconnaissance, social engineering and support for malicious tools. Microsoft’s new assessment expands on this trend with specific examples and the claim that AI is a routine tool among several tracked groups. However, Microsoft does not report confirmation of autonomous execution of offensive steps in this case.

Verified and updated: 10/30/2026 06:27