Secure AI Deployment Checklist
An interactive checklist organized by OWASP LLM Top 10 2025 for evaluating and deploying AI systems with security, privacy, and governance built in from day one.
Resources
Tools & Reading
Guides, interactive tools, and curated reading for security leaders navigating AI risk. Built from what I actually use and recommend.
Downloadable Guides
Frameworks & Checklists
Interactive Tools
From the Lab
Reading List
What I'm Reading
Papers, frameworks, and posts that have shaped how I think about AI security. Each annotation explains why it matters.
- Responsible Scaling Policy v3.0↗
Anthropic
The clearest framework I've seen for tying AI capability thresholds to concrete safety commitments. Essential reading for anyone writing governance policy around frontier models.
- MITRE ATLAS: Adversarial Threat Landscape for AI Systems↗
MITRE Corporation
ATLAS does for ML what ATT&CK did for enterprise security. It provides a shared taxonomy of adversarial tactics against AI systems. Invaluable for threat modeling AI deployments.
- AI Risk Management Framework Playbook (AI RMF 1.0)↗
National Institute of Standards and Technology
Practical companion to the NIST AI RMF that translates abstract governance principles into actionable suggested actions. I reference the Map and Measure functions regularly.
- OWASP Top 10 for Large Language Model Applications (2025)↗
OWASP Foundation
The definitive checklist for LLM-specific vulnerabilities, from prompt injection to training data poisoning. A must-have for security reviews of any LLM integration.
- Guidelines for Secure AI System Development↗
UK NCSC, CISA, and International Partners
Joint guidance from 18 agencies across 6 countries. The most authoritative multi-stakeholder baseline for secure-by-design AI development I've encountered.
- Securing the AI Software Supply Chain↗
Trail of Bits
Goes beyond model security to address the full pipeline: data provenance, dependency risks, and CI/CD for ML. Bridges the gap between traditional AppSec and ML engineering.
- Sleeper Agents: Training Deceptive LLMs That Persist Through Safety Training↗
Hubinger et al. (Anthropic)
Demonstrates that standard safety training may not remove deceptive behaviors once established. Changed how I think about trust boundaries for fine-tuned models.