Actor Profile

Storm-3168 is Microsoft's tracking designation for JADEPUFFER, a threat actor first documented by Sysdig in July 2026 as the first known agentic ransomware operation. The actor demonstrates advanced cloud-focused capabilities, leveraging AI-orchestrated techniques to conduct complex post-compromise operations across Azure environments with speed and scale. Storm-3168 targets cloud infrastructure through compromised service principals, conducting extensive reconnaissance, bulk resource destruction, and credential harvesting operations. The actor's evolution from initial documentation to Azure-focused attacks represents a broader shift toward AI-driven threat operations in cloud environments.

TTPs (Tactics, Techniques, Procedures)

Storm-3168 employs multi-stage cloud attack techniques beginning with initial access via exposed credentials (T1552.001 - Credentials In Files) from public GitHub repositories containing plaintext service principal secrets. Discovery operations include extensive enumeration of Azure resources (T1580 - Cloud Infrastructure Discovery, T1526 - Cloud Service Discovery) targeting virtual machines, subscriptions, resource groups, storage accounts, and App Service configurations. The actor conducts bulk destructive operations (T1485 - Data Destruction, T1489 - Service Stop) targeting Azure Storage Accounts, SQL databases, Key Vaults, Function Apps, Virtual Machines, and App Services, with 100+ deletion attempts executed within a 7-minute window. Credential access activities (T1552.005 - Cloud Instance Metadata API) include 30+ successful ListKeys API requests to harvest storage account access keys. The actor uses python-requests/2.34.2 user agent and demonstrates parallel execution capabilities across multiple resource types simultaneously.

Targets & Patterns

Storm-3168 primarily targets cloud services and enterprise IT sectors, with specific focus on Azure cloud environments. The actor targets organizations with significant Azure footprints, seeking to compromise service principals with broad administrative permissions across subscriptions and resource groups. Attack patterns indicate the actor conducts comprehensive reconnaissance across entire Azure tenants before executing destructive operations, suggesting targeting of organizations with valuable cloud-hosted data and critical infrastructure dependencies. The actor also probes Azure App Services across multiple customers, targeting sensitive paths related to WordPress administration and PHP-CGI, indicating opportunistic scanning for vulnerable web applications. The focus on Azure Site Recovery storage accounts and backup protection locks suggests intent to prevent recovery operations and maximize operational impact on victim organizations.

Historical Context

JADEPUFFER was first documented by Sysdig in July 2026 as the first known agentic ransomware operation, representing a significant evolution in threat actor capabilities through AI orchestration. Microsoft's investigation expands the publicly documented activity associated with JADEPUFFER, providing the first detailed analysis of Storm-3168's Azure-specific operations. The observed activity in early June 2026 demonstrates the actor's evolution from initial documentation to sophisticated multi-stage cloud attacks. This campaign highlights a broader industry shift toward AI-orchestrated attacks where threat actors coordinate complex post-compromise operations with greater speed and scale than traditional manual operations. The activity represents an escalation in cloud-focused destructive capabilities, with the actor demonstrating ability to execute 150+ malicious operations within 35 minutes across compromised Azure environments.

Defensive Recommendations

  • Immediately revoke and rotate any service principal credentials exposed in public repositories, including GitHub issue histories; removal of exposed secrets does not invalidate them and they remain usable until explicitly revoked
  • Implement Azure resource locks and storage account-level deletion protection to prevent destructive operations even when identities with administrative permissions are compromised
  • Enforce least privilege access for service principals and workload identities, limiting permissions to only required operations and specific resource scopes to reduce blast radius of compromise
  • Enable Microsoft Defender for Cloud protections and monitor for anomalous Azure Resource Manager (ARM) API activity including bulk ListKeys operations (T1552.005), rapid resource enumeration (T1580), and parallel deletion attempts (T1485)
  • Implement detection rules for suspicious user agent strings (e.g., python-requests/2.34.2) combined with high-volume API operations, and alert on attempts to disable Azure Site Recovery locks or Azure Backup protection mechanisms
  • Scan public code repositories and issue trackers for exposed Azure credentials (client IDs, client secrets, tenant IDs) and implement automated secret scanning tools to prevent credential leakage