StrongDM · Agent Security Charter

Securing theAI Agent Lifecycle

OWASP Top 10 for Agentic Applications. Enterprise-grade governance with Leash. Identity, authorization, and trust for the agentic era.

Based on OWASP Top 10 for Agentic Applications 2026
OWASP Top 10 for Agentic Applications

Understanding the Threats

AI agents introduce novel attack surfaces. The OWASP Top 10 for Agentic Applications defines the critical security risks organizations must address.

🎯
ASI01
critical

Agent Goal Hijack

Attackers manipulate an agent's objectives through prompt injection, causing it to pursue malicious goals while appearing to operate normally.

Prompt injection via user inputIndirect injection through retrieved documents+1 more
View details & mitigations
🔧
ASI02
critical

Tool Misuse and Exploitation

Agents use their available tools in unintended ways, either through manipulation or emergent behavior, leading to unauthorized actions.

Chaining tools to bypass restrictionsUsing file tools to access sensitive data+1 more
View details & mitigations
🔐
ASI03
critical

Identity and Privilege Abuse

Agents operate with excessive permissions or impersonate users, leading to unauthorized access to systems and data.

Agents inheriting user permissionsCross-tenant data access+1 more
View details & mitigations
📦
ASI04
high

Agentic Supply Chain Vulnerabilities

Compromised components in the agent stack—models, tools, plugins, or dependencies—introduce security risks.

Malicious MCP serversBackdoored tool libraries+1 more
View details & mitigations
StrongDM Solution

Leash: Agent Governance

A multi-layer governance system that combines kernel-level enforcement with application-layer intelligence. Record, shadow, and enforce agent behavior.

Kernel Layer
🔒

eBPF LSM Kernel Hooks

Intercept file operations, process execution, and network connections at the kernel level with zero agent modification.

Proxy Layer
🌐

HTTPS MITM Proxy

Observe and enforce policies on HTTP traffic, API calls, and MCP tool invocations with semantic understanding.

Policy Engine
📜

Cedar Policy Language

Express complex access control policies in a readable, auditable language designed for authorization decisions.

Operational
📊

Record → Shadow → Enforce

Progressive deployment model: observe behavior, validate policies in shadow mode, then enforce with confidence.

Interactive Tools

Explore the Charter