CrowdStrike CNAPP Enhancements Prioritize Adversary-Informed Cloud Risks
CrowdStrike's new CNAPP capabilities enhance cloud risk assessment by incorporating application-layer visibility, adversary intelligence, and configuration change tracking, enabling security teams to prioritize remediation based on real-world threat actor behavior such as LABYRINTH CHOLLIMA and SCATTERED SPIDER.
Triage facts
Revision 2- Evidence
- Unknown
- Severity
- High
- Sources
- 1 public source
- Freshness
- Revision 2Mar 30, 09:13 UTC
- Detection
- 2 rules
Revision history 1 entry · Revision 2
- Revision 2
Repaired a legacy canonical identity collision
Changed: slug
CrowdStrike has enhanced its Cloud Native Application Protection Platform (CNAPP) with new features designed to address limitations in current cloud risk assessment methodologies. These enhancements, released in March 2026, focus on providing comprehensive visibility, intelligent risk prioritization, and actionable insights. The key features include Application Explorer, which provides application-layer visibility, and adversary intelligence integration, which prioritizes risks based on the observed behavior of threat actors like LABYRINTH CHOLLIMA and SCATTERED SPIDER. The enhancements aim to reduce alert fatigue and enable security teams to focus on the most critical risks that align with real-world threat actor tactics, techniques, and procedures (TTPs). According to the CrowdStrike 2026 Global Threat Report, cloud-conscious intrusions by state-nexus threat actors surged 266% year-over-year in 2025, highlighting the increasing importance of proactive cloud security measures.
Attack Chain
- Initial Access: An attacker identifies a cloud environment with misconfigured storage resources through reconnaissance.
- Credential Access: The attacker exploits overly permissive access controls on a storage resource to gain access to stored credentials.
- Discovery: The attacker uses the compromised credentials to enumerate applications connected to the storage resource and identifies those processing sensitive data, such as customer PII.
- Lateral Movement: The attacker leverages application dependencies to move laterally to other cloud services and resources within the environment.
- Collection: The attacker targets AI-driven applications to discover shadow AI activity and identifies dependencies on external large language models (LLMs).
- Exfiltration: The attacker exfiltrates sensitive data, including customer PII, by exploiting unapproved model usage and exposing data to external AI services.
- Impact: The attacker achieves their objective by accessing and potentially selling or using the exfiltrated sensitive data for malicious purposes.
Impact
The enhanced CNAPP capabilities are designed to mitigate the increasing risk of cloud breaches, which saw a 266% surge in intrusions by state-nexus actors in 2025. If attackers successfully exploit cloud misconfigurations and vulnerabilities, organizations could experience data breaches, financial losses, reputational damage, and regulatory penalties. The integration of adversary intelligence helps organizations prioritize and address risks that align with the documented TTPs of known threat actors, reducing the likelihood of successful attacks.
Recommendation
- Deploy the Sigma rule for detecting overly permissive storage access to identify potential initial access points, focusing on cloud environments (AWS S3 buckets, Azure Blob Storage, Google Cloud Storage) and tune for your environment.
- Implement the Sigma rule for identifying shadow AI activity and unauthorized LLM usage to prevent sensitive data exposure to external AI services.
- Utilize Falcon Cloud Security's Application Explorer to gain visibility into application dependencies and identify potential lateral movement paths.
- Prioritize remediation efforts based on the adversary intelligence provided by Falcon Cloud Security, focusing on risks aligned with the TTPs of threat actors known to target your industry.
Detection coverage 2
Detect Overly Permissive Cloud Storage Access
highDetects overly permissive access configurations on cloud storage services, which can be exploited for initial access and data exfiltration.
Detect Shadow AI LLM Usage
mediumDetects applications using external Large Language Models (LLMs) that may not be approved or monitored, leading to potential data exposure.
Detection queries are available on the platform. Get full rules →