Microsoft says North Korean threat actors adopt AI unevenly across operations

On Oct. 29, 2026, Microsoft Threat Intelligence said North Korean threat actors are broadening AI use at uneven adoption rates and applying the technology as a force multiplier to speed up and scale existing workflows.

Microsoft rates AI adoption across North Korean threat actors as high, noting that Jade Sleet, Coral Sleet, and Jasper Sleet have put AI into use throughout their full operations. The company also judges that people continue to direct AI‑enabled operations by these actors, and the operations have not turned fully autonomous. Within that high rating, Microsoft places Diamond Sleet, Onyx Sleet, Pearl Sleet, and Sapphire Sleet at the accelerating stage of AI adoption, while Emerald Sleet sits at the expanding stage; the assessment finds no North Korean threat actors at the emerging stage. AI adoption refers to the extent to which a group uses artificial intelligence in its operations.

Jade Sleet illustrates the faster‑moving end of the adoption curve. In one week the actor ran campaigns using three front companies, and from June 8 through June 12, 2026, it carried out job‑themed targeting while posing as recruiters for a fake AI‑powered blockchain intelligence company. Front companies are fake businesses used to disguise the true operator.

Coral Sleet quickly created a new ability to trojanize private source code repositories. Since April 2026 the group has operated a workflow for trojanizing private GitHub repositories, a capability Microsoft judges as highly likely engineered through an AI‑enabled development platform. Trojanizing involves inserting malicious code into legitimate software.

Jasper Sleet uses generative AI to construct fake digital personas and deploys fraudulent identities to land legitimate remote jobs. The actor relies on AI tooling to keep messaging and performance steady across many fake worker identities. Generative AI refers to systems that can create new content such as text or images.

Diamond Sleet applied AI to improve its targeting processes, covering persona creation and the setup of front companies. Since August 2026 the group returned to oil and gas targeting for the first time since 2023 and created fake communications technology companies to target telecommunications companies. Persona creation means building a detailed background for a fictional identity.

Onyx Sleet produced dozens of customized lures within minutes. Microsoft identified over 100 PDF lure files attributed to Onyx Sleet created in August and September 2026, with dozens produced at a time within minutes. Lures are deceptive files or messages designed to trick recipients.

Sapphire Sleet likely incorporated AI during the August 2026 Rust software compromise. In that campaign the actor’s malware used dynamic .NET compilation during post‑exploitation, most likely to evade defenses. Post‑exploitation refers to actions taken after an attacker has gained access to a system.

Since late October 2026, Onyx Sleet and Pearl Sleet have leveraged the open‑source agentic AI penetration testing framework ARTEX in operations. Agentic AI tools act with a degree of autonomy rather than generating static content.

Microsoft identified possible synthetic media generation focused on AI‑created voice and video by Emerald Sleet and assesses it as highly likely that many North Korean threat actors value synthetic media creation for enhancing social engineering. Synthetic media is content generated by AI, such as voice or video.

Beyond individual actors, North Korean threat actors use AI for reconnaissance, researching organizations, employees, wallets, exchanges, decentralized finance, blockchain workflows, vulnerabilities, APIs, security products, and cloud services. Reconnaissance is the process of gathering information about a target.

The technology also powers the creation of tailored phishing, recruiter, and investment lures, as well as job‑themed payload‑delivery content and malicious repository lures. Phishing is a technique that uses deceptive messages to steal credentials.

Actors use AI to translate content and adjust tone, to build resumes, biographies, and realistic interview responses, and to enhance and alter profile images, video, voice, and backgrounds. These capabilities sustain recruiter, developer, investor, and other social identities across interactions and platforms.

North Korean threat actors use AI‑assisted development platforms such as Cursor to generate, convert, debug, and refactor code, assist with scripts, commands, persistence, and obfuscation, and troubleshoot errors. AI‑enabled development cycles help actors circumvent detection by streamlining retooling and enabling rapid pivots in reaction to defensive measures. Obfuscation hides the true purpose of code, while persistence maintains access to a compromised system.

North Korean threat actors run cyber operations mainly to fund the regime, gather strategic intelligence, and occasionally pursue politically motivated disruption or destruction. Common targets include cryptocurrency and financial firms, information technology companies and their software supply chains, government bodies, and education‑related organizations. North Korean threat actors mostly adopt AI to back current goals and targeting priorities instead of changing their underlying motivations. Extensive international sanctions constrain the Democratic People’s Republic of Korea, which turns to cyber operations and other illicit methods to bypass restrictions and advance state goals.

Microsoft introduced the Security Dashboard for AI, which aggregates security, identity, and data risk across Microsoft Defender, Microsoft Entra, and Microsoft Purview. Microsoft Agent 365 acts as a control plane for AI agents, letting users manage, govern, and secure agents and workflows and watch for emerging risks tied to agentic AI.

The pattern fits a wider trend documented in Microsoft’s 2026 Digital Defense Report. The report says attackers gain from AI more quickly than defenders, accelerating vulnerability discovery, malware development, and activity after a compromise. AI cuts the time, skill, and spending needed to find and exploit flaws, and the median window from wild vulnerability discovery to weaponization has dropped well under 24 hours.

Remediation is inherently slower than discovery, and Microsoft warns of a likely multi‑year period in which the number of known but unpatched vulnerabilities spikes. AI helps attackers build custom malware and speed up post‑compromise steps such as data theft, credential hunting, and lateral movement, compressing them from days into minutes. Remediation means fixing a vulnerability after disclosure.

Nation‑state threat actors, including Chinese, Russian, and North Korean groups, have started using AI in real‑world operations for research, malware development, social engineering, and other attack components. North Korean remote IT workers apply AI to persona development, social engineering, and keeping access to organizations, whereas other North Korean actors use it to build malware and run attack infrastructure. A nation‑state actor is a hacking group directed or funded by a government.

“For sophisticated actors, AI allows unprecedented speed, scale, and customization, reducing the attack chain from days to seconds,” explains Microsoft. The company adds that defenders will eventually gain similar benefits from AI, but attackers currently hold the advantage.