Tracking AI adoption in North Korean threat actor tradecraft | Microsoft Security Blog
Eight named North Korean threat actor groups are integrating AI into operational workflows to accelerate phishing, persona creation, malware development, and software supply chain attacks. Multiple North Korean threat actor groups have integrated AI into workflows spanning reconnaissance, social engineering, malware development, and supply chain compromise. Jade Sleet, Coral Sleet, and Jasper Sleet have operationalized AI across entire operations, while Onyx Sleet and Pearl Sleet leverage the ARTEX agentic AI framework for penetration testing.
Improve this report
Send corrections, useful missing information, or quality feedback. It is private and will not appear as a comment.
Detection / HunterZ.ai
Immediate Actions
- Block device code flow authentication unless explicitly required by policy
- Enable EDR in block mode and cloud-delivered protection on all endpoints
- Review and purge recently delivered job-themed phishing emails using Threat Explorer
Key Takeaways
- Jade Sleet, Coral Sleet, and Jasper Sleet have operationalized AI across all operational workflows
- Onyx Sleet and Pearl Sleet leverage ARTEX, an open-source agentic AI penetration testing framework
- Onyx Sleet created over 100 target-specific PDF lures in August-September 2026, dozens within minutes
- Coral Sleet built a new capability to trojanize private GitHub repositories since April 2026
- Sapphire Sleet's August 2026 Rust compromise used dynamic .NET compilation for defense evasion
Attack Chain
- Reconnaissance: AI accelerates research of target organizations, employees, wallets, and technology stacks
- Social Engineering: AI generates tailored PDF lures, fake personas, and job-themed content at scale
- Initial Access: Malicious PDF attachments or trojanized GitHub repositories deliver malware to victims
- Execution: Malware deployed with AI-assisted obfuscation and dynamic .NET compilation
- Defense Evasion: AI platforms enable rapid retooling to circumvent detection and blocking measures
- Persistence: AI assists development of persistence mechanisms and management of fraudulent worker personas
Affected Systems
- Rust software supply chain ecosystems
- Private GitHub repositories
- Cryptocurrency and financial technology organizations
- IT companies and software supply chains
- Government and defense sector organizations
- Telecommunications companies
Detection Availability
- Platforms: Microsoft Defender for Endpoint, Microsoft Defender for Office 365, Microsoft Entra ID Protection, Microsoft Defender for Cloud
Microsoft Defender detection names and alert coverage listed for multiple North Korean threat actors across tactics.
Detection Engineering Assessment
Hunting Hypotheses
| Hypothesis | Telemetry | Stage | FP Risk |
|---|---|---|---|
| Rapid sequential creation of similar PDF documents with metadata timestamps within minutes indicates automated lure generation (T1566.002) | File creation events with metadata timestamps | Initial Access | Medium — legitimate marketing teams batch-generate content |
| Newly created social media profiles with enhanced or altered images contacting employees about job opportunities indicates AI-assisted persona operations (T1585) | Social media intelligence and email gateway logs | Reconnaissance | High — legitimate recruiters use new profiles |
| Dynamic .NET compilation activity during post-exploitation indicates defense evasion by Sapphire Sleet (T1027) | Process and assembly load events | Defense Evasion | Low — dynamic compilation in post-exploitation is rare |
| GitHub repository forks or clones followed by malware deployment indicates trojanized supply chain activity (T1195.002) | GitHub audit logs and endpoint process telemetry | Execution | Medium — CI/CD pipelines may trigger similar patterns |
Required Log Sources
- Microsoft Entra ID sign-in logs
- Microsoft Defender for Office 365 email logs
- Process creation logs (Sysmon EID 1)
- GitHub repository audit logs
- Microsoft Defender for Cloud alerts
Control Gaps
- AI-generated social engineering lures evade traditional email security
- Fraudulent remote worker identities bypass standard hiring vetting
- Trojanized private repositories bypass code review controls
- Device code flow abuse bypasses MFA enforcement
- Synthetic media generation evades identity verification checks
Recommendations
Infrastructure Hardening
- Isolate web-facing systems behind VPN, WAF, or DMZ perimeter networks
- Enable attack surface reduction rules in Microsoft Defender XDR
- Automate patching using centralized update management across cloud and on-premises
- Enable Zero-hour auto purge in Defender for Office 365 to quarantine malicious mail
User Protection
- Enable Safe Links and Safe Attachments in Defender for Office 365
- Train staff to identify job-themed phishing and AI-generated lure content
- Verify remote worker identities through multiple video sessions with cameras on