Snyk Agent Scan, NIST AI Control Overlays, and the CSA AI Controls Matrix#

As AI agents move from experimental to production deployments, three critical developments are reshaping how organizations secure and govern them: Snyk’s open-source agent-scan tool addresses the immediate need for agent security scanning, while NIST and CSA provide the governance frameworks that organizations need for compliance and risk management.

Snyk Agent Scan#

Snyk agent-scan is an open-source security scanner specifically designed for AI agents, MCP servers, and agent skills. It addresses a critical gap — most security scanning tools were built for traditional software, not for AI agent configurations.

What It Scans#

agent-scan targets three key areas:

1. Agent Configurations

Auto-discovers local agent configurations (Claude, Cursor, Windsurf, Gemini CLI)
Analyzes system prompts, tool permissions, and access controls
Identifies overly permissive configurations

2. MCP Servers

Connects to MCP servers to fetch tool descriptions
Scans tool definitions for hidden instructions (tool poisoning)
Validates MCP server configurations against security best practices

3. Agent Skills

Scans skill files for malicious payloads and prompt injections
Checks for insecure configurations and leaked secrets
Validates against known malicious patterns

The Skill Inspector#

Snyk also provides a Skill Inspector — a free, self-service web tool that instantly scans agent skills for security issues:

Detects malicious skills before installation
Identifies insecure configurations
Finds leaked secrets embedded in skill definitions
Provides actionable remediation guidance

The CLI-Anything Threat#

Snyk’s research revealed a significant supply chain risk: CLI-Anything — a technique where a single command turns any open-source repository into an AI agent backdoor by generating SKILL.md files that AI agents trust and execute automatically. Their analysis found that 13.4% of agent skills contain critical security issues, and no traditional SAST or SCA scanner detects malicious instructions in these files.

1
# Install and run agent-scan
2
npm install -g @snyk/agent-scan
3
agent-scan scan --path ./my-agent-project
4

5
# Scan a specific skill
6
agent-scan skill-inspector --skill ./skills/my-skill.md
7

8
# Scan MCP server configurations
9
agent-scan mcp --config ./mcp-config.json

Snyk + Vercel Partnership#

Snyk has partnered with Vercel to secure the skills.sh ecosystem, integrating real-time security scanning to detect malicious payloads and prompt injections before they reach developers’ machines.

NIST SP 800-53 Control Overlays for AI Systems#

NIST Control Overlays for Securing AI Systems represents a significant evolution in AI governance — adapting the widely-used SP 800-53 security controls specifically for AI.

What Are Control Overlays?#

Control overlays are tailored sets of security controls that adapt, supplement, and refine the baseline SP 800-53 controls for specific technologies. The AI overlays:

Map AI-specific risks to existing security controls
Add new controls unique to AI systems
Provide implementation guidance specific to AI deployments
Support federal compliance requirements for AI systems

Key Areas Covered#

Domain	Overlay Focus
Data Governance	Training data integrity, provenance, and quality controls
Model Security	Model integrity, access control, and monitoring
Inference Security	Input validation, output filtering, rate limiting
Agent Security	Tool permissions, scope limitation, audit logging
Supply Chain	Model provenance, dependency verification, artifact scanning
Privacy	Data minimization, differential privacy, consent management
Bias & Fairness	Testing for discriminatory outputs, fairness metrics

Timeline#

August 2025: Concept paper and proposed action plan released
January 2026: Annotated outline (discussion draft) published
Ongoing: Community engagement through workshops and public comment

NIST AI Agent Standards Initiative#

In February 2026, NIST’s Center for AI Standards and Innovation (CAISI) formally launched the AI Agent Standards Initiative — the first US government program dedicated exclusively to establishing security and safety standards for AI agents. This initiative will produce:

Agent identity and authentication standards
Agent behavior monitoring requirements
Agent-to-agent communication security standards
Agent deployment and operational controls

CSA AI Controls Matrix#

The Cloud Security Alliance AI Controls Matrix (AICM) is a vendor-neutral framework with 243 controls across 18 domains for assessing and managing AI-specific security and trust risks.

Structure#

The AICM is aligned with major frameworks:

ISO/IEC 42001: AI Management System requirements
NIST AI RMF: Risk management framework
CSA Cloud Controls Matrix (CCM): Existing cloud security controls
EU AI Act: European AI regulation requirements

The 18 Domains#

The matrix covers controls across these domains:

AI Governance and Risk Management
Data Quality and Integrity
Model Development and Training
Model Validation and Testing
Deployment and Operations
Monitoring and Observability
Incident Response
Access Control and Identity
Privacy and Data Protection
Bias and Fairness
Transparency and Explainability
Supply Chain Security
Agent Security
Infrastructure Security
Compliance and Audit
Business Continuity
Change Management
Third-Party Management

Using the AICM#

1
Step 1: Assess current AI deployments against AICM controls
2
Step 2: Identify gaps between current state and AICM requirements
3
Step 3: Prioritize remediation based on risk and business impact
4
Step 4: Implement missing controls
5
Step 5: Validate implementation through testing
6
Step 6: Maintain continuous compliance through monitoring

Building a Comprehensive AI Governance Stack#

Layer 1: Standards and Frameworks#

1
NIST AI RMF → Risk management methodology
2
NIST SP 800-53 AI Overlays → Security control requirements
3
CSA AICM → Detailed control specifications
4
ISO/IEC 42001 → Certifiable management system

Layer 2: Technical Controls#

1
Snyk agent-scan → Agent and skill scanning
2
modelscan → Model artifact scanning
3
MCP-Scan → MCP server scanning
4
llm-guard → Runtime guardrails

Layer 3: Continuous Monitoring#

1
Snyk Agent Security (Evo AI-SPM) → Full lifecycle governance
2
CSA MCP Security Resource Center → MCP-specific guidance
3
NIST CAISI standards → Agent security standards

Key Takeaways#

Snyk’s agent-scan fills a critical gap by scanning AI agent configurations, MCP servers, and skills — areas traditional security tools do not cover
The CLI-Anything threat reveals that 13.4% of agent skills contain critical security issues, highlighting the supply chain risk
NIST’s AI Control Overlays bring the rigor of SP 800-53 to AI systems, providing a path for federal AI compliance
The CSA AI Controls Matrix with 243 controls across 18 domains provides the most detailed control framework available for AI security
Organizations should align their AI security programs with both NIST overlays and CSA AICM for comprehensive governance coverage

Newsletter