AI agents introduce security problems that traditional tools weren't built for. These guides explain the threats, the defenses, and where the gaps are.
Get Started
Set up Pipelock with your IDE in minutes.
Claude Code
One-command hook setup for Bash, WebFetch, Write, Edit, and MCP scanning. The fastest path to agent security.
Cursor IDE
One-command install. DLP and injection blocking for every agent action in Cursor.
VS Code
MCP proxy integration for VS Code. Wraps all MCP servers through Pipelock’s scanner.
JetBrains / Junie
Wrap Junie MCP servers with bidirectional scanning. IntelliJ, PyCharm, WebStorm, GoLand.
Community Rules
Install signed, versioned detection patterns with one command. Hot-reloadable.
Understand the Threats
Deep dives on attack vectors and defenses.
MCP Security
Tool poisoning, rug-pulls, and how to protect MCP connections. Maps to OWASP MCP Top 10.
Agent Egress Security
Preventing credential leaks and data exfiltration from AI agents via URLs, headers, and bodies.
Prompt Injection Defense
Catching injection at the proxy layer with 6-pass normalization, not just at the model.
Canary Tokens
Plant synthetic credentials in your environment. Catch real exfiltration regardless of encoding.
Evidence and Compliance
Prove your agent security posture to auditors, customers, and regulators.
Flight Recorder
Hash-chained, tamper-evident audit log with Ed25519 signed checkpoints and redaction support.
Compliance Evidence
OWASP, NIST 800-53, EU AI Act, SOC 2 framework mappings. Signed assessment reports.
EU AI Act Compliance
How Pipelock’s runtime controls map to EU AI Act requirements for high-risk AI systems.
Security Framework Coverage
How Pipelock maps to established security frameworks.
OWASP Agentic Top 10
10/10 categories covered. Per-threat assessment with specific scanner mappings.
OWASP AIVSS Coverage
Maps to all 10 agentic risk categories in the AIVSS v0.8 scoring system.
OWASP Top 10 for LLMs
7/10 categories covered. Network-layer defense for the 2025 LLM Top 10.
OWASP Agentic AI Threats
12/15 threats covered in the Agentic AI framework.
SlowMist MCP Security
10 full, 8 partial coverage against the 19-item MCP security validation guide.