JADEPUFFER Ransomware Erases Cloud Infrastructure in Seven Minute Attack

The ransomware landscape has crossed a new threshold in late 2026. Microsoft Threat Intelligence detailed an attack by the group it tracks as JADEPUFFER (Storm 3168), an AI-driven ransomware operation that wiped more than 100 Azure storage accounts in just seven minutes. The attack did not encrypt files in the traditional sense. Instead, it destroyed cloud infrastructure outright, targeting storage accounts, key vaults, function apps, virtual machines, and app services with surgical precision.

The Attack Anatomy

What makes JADEPUFFER particularly alarming is its use of two compromised Azure service principals whose credentials had been leaked in a public GitHub issue. Rather than using a single identity for the entire attack, the operation employed a division of labor that mirrors professional software development practices:

• First service principal: Spent 15.5 hours executing over 300 read operations across subscriptions and virtual machines, mapping the entire cloud environment
• Second service principal: Performed 30 minutes of discovery followed by 150+ destructive and credential operations over 35 minutes

The most destructive phase — the deletion of 100+ storage accounts, a key vault, a function app, an app service plan, SQL databases, and virtual machines — was completed in approximately seven minutes. Some deletion attempts were blocked by Azure resource locks (CanNotDelete), which saved a portion of the infrastructure. However, the attacker also fired 30+ ListKeys requests targeting storage and Azure Site Recovery keys, attempting to compromise recovery mechanisms.

From Langflow Exploit to Cloud Devastation

JADEPUFFER first appeared in July 2026 as an agentic ransomware operation exploiting CVE-2025-3248, a vulnerability in the Langflow AI workflow platform. This origin story is significant because it demonstrates how ransomware groups are increasingly targeting AI infrastructure as an entry point. The group then pivoted to exploiting leaked cloud credentials, showing an adaptive approach that shifts tactics based on available access vectors.

Microsoft could not confirm the initial access route for the June attacks, but confirmed that credentials for one service principal had appeared in a public GitHub issue. This detail underscores a persistent problem: developers and administrators continue to accidentally expose cloud credentials in code repositories, configuration files, and issue trackers, providing ransomware operators with ready-made access to enterprise cloud environments.

The Broader AI Ransomware Ecosystem

JADEPUFFER does not exist in isolation. Security researchers simultaneously disclosed the CLOSEDQUORUM Windows implant, which allows four large language models to vote on post-compromise actions. This represents a fundamentally new paradigm where AI models collectively decide which lateral movement techniques to use, which credentials to target, and which systems to destroy.

The convergence of these developments signals that ransomware operators are no longer simply deploying encryption payloads. They are building autonomous attack systems that can:

• Reconnoiter cloud environments using read operations that blend into normal administrative activity
• Execute destructive actions faster than human incident responders can react
• Target backup and recovery infrastructure to eliminate restoration options
• Adapt tactics based on what they encounter in real time

Cloud Security Implications

The JADEPUFFER attack exposes several critical gaps in how organizations secure their cloud infrastructure. Traditional ransomware defense focused on endpoint protection and network segmentation. Cloud-native ransomware requires a fundamentally different approach:

Service Principal Governance

Service principals are powerful non-human identities that can access resources across an entire cloud tenant. When their credentials are leaked, they provide attackers with persistent, high-privilege access that bypasses multi-factor authentication and other user-centric controls. Organizations must implement credential rotation policies, monitor for leaked secrets in public repositories using tools like GitHub Advanced Security, and enforce least-privilege access on every service principal.

Resource Locks as Last-Line Defense

In the JADEPUFFER attack, CanNotDelete resource locks prevented some deletion attempts, preserving portions of the infrastructure. This demonstrates that resource locks, often overlooked as a basic governance tool, can serve as a critical defensive layer against destructive cloud attacks. Every storage account, key vault, and critical resource should have appropriate locks applied.

Alerting on Destructive Operations

The seven-minute destruction window means that alert-driven response is insufficient. By the time a security team receives and triages an alert about mass resource deletion, the damage is already done. Organizations need automated response capabilities — Azure Logic Apps, Sentinel playbooks, or equivalent — that can immediately disable compromised service principals and quarantine resources when anomalous destructive activity is detected.

The Edge Appliance Vulnerability Wave

JADEPUFFER’s cloud-native attack coincides with a broader wave of zero-day exploits targeting edge appliances that serve as gateways to enterprise networks. Unauthenticated remote code execution is actively occurring against Citrix NetScaler ADC and Gateway appliances via CVE-2026-88771 and CVE-2026-88772, both rated 9.5 Critical. With over 50,000 potentially vulnerable internet-facing NetScaler instances identified globally, this vulnerability provides ransomware groups with a massive attack surface.

Simultaneously, CISA added CVE-2026-63077 to its Known Exploited Vulnerabilities catalog — a critical authentication bypass in JetBrains TeamCity that ransomware gangs are actively abusing. This is the fourth TeamCity vulnerability CISA has linked to ransomware since 2023. The pattern is clear: ransomware groups systematically target continuous integration and deployment infrastructure as a reliable entry point into enterprise environments.

Defensive Recommendations

Based on the JADEPUFFER attack pattern and the broader threat landscape, organizations should take the following actions:

• Audit all service principals and managed identities for excessive permissions. Remove unused credentials and enforce rotation schedules.
• Scan public repositories for leaked credentials using automated secret scanning tools. Treat every exposed credential as a confirmed breach.
• Apply CanNotDelete locks to all critical cloud resources, including storage accounts, key vaults, and recovery services.
• Deploy automated response playbooks that disable compromised identities and isolate resources within seconds of detecting destructive activity.
• Patch internet-facing infrastructure immediately, particularly Citrix NetScaler, JetBrains TeamCity, Check Point VPN, and Roundcube installations.
• Monitor for anomalous read operations — the 15.5-hour reconnaissance phase is a detection opportunity that many organizations miss.

The New Normal

JADEPUFFER represents a shift from encryption-based extortion to pure infrastructure destruction. When attackers can wipe 100+ storage accounts in seven minutes using leaked credentials and AI-driven decision-making, the traditional backup-and-restore model is no longer sufficient. Cloud resilience now requires proactive identity governance, automated threat response, and a fundamental assumption that any exposed credential will eventually be weaponized.

The ransomware ecosystem has evolved from manual operations run by human attackers to autonomous systems that can map, decide, and destroy faster than any incident response team can react. Organizations that continue to treat cloud security as an extension of perimeter defense will find themselves in JADEPUFFER’s seven-minute window — with no time to respond and no infrastructure left to recover.

---

Edited by Palawan @QUE.COM
Website: https://QUE.COM Intelligence
Sponsored by: https://MAJ.COM AI Autonomous