# AI Defense Stack

> AI security tools, compared on verified facts. We compare 111 AI security tools (52 open source) on facts checked at each tool's own primary source: every fact shows its source URL and the date we checked it. We compile from public sources and do not run the products. We score tools with a published rubric (the Coverage Score) and map them to the OWASP Top 10 for LLM Applications 2025 and MITRE ATLAS. The mapping is our analysis.

Key facts for citation:
- Scope: seven tool categories, plus courses and certifications.
- 111 tool listings, 12 courses and certifications, 11 guides, 8 comparisons, 57 glossary terms, 60 answered questions.
- Every factual field is a Fact: a value, a source_url and an accessed date. Our own analysis is labeled "Our analysis".
- Independent: sponsorship and affiliate relationships never change scores, rankings or verdicts, and each is disclosed.
- Publisher: AI Defense Stack team. Content is drafted with AI assistance and facts are checked against primary sources.
- Contact: contact@aidefensestack.com
- Last updated: 2026-10-21.

## Categories
- [LLM firewalls & guardrails](https://aidefensestack.com/categories/guardrails/): 26 tools (1 cross-listed). Checks that screen what goes into and comes out of a model: libraries, safety models, cloud APIs and gateways.
- [AI red teaming & LLM evaluation](https://aidefensestack.com/categories/red-teaming/): 24 tools. Scanners, attack frameworks, evaluation harnesses and managed services that test models and LLM apps on purpose.
- [Model & ML supply-chain security](https://aidefensestack.com/categories/model-supply-chain/): 10 tools (1 cross-listed). Scanners, signing tools and registry controls that check models and ML artifacts before they run.
- [AI security posture management & discovery](https://aidefensestack.com/categories/ai-spm/): 10 tools (2 cross-listed). Tools that find AI models, services and data in your environments and check them for risky settings.
- [Agent & MCP security](https://aidefensestack.com/categories/agent-security/): 29 tools (1 cross-listed). Scanners, gateways and runtime controls for AI agents that call tools through the Model Context Protocol.
- [GenAI data-leak prevention & privacy](https://aidefensestack.com/categories/genai-data-protection/): 10 tools. Tools that keep sensitive data out of AI tools and models, or remove it first.
- [AI governance, risk & compliance](https://aidefensestack.com/categories/ai-governance/): 7 tools. Platforms that list AI systems, assess their risks, manage policies and keep evidence for audits.
- [Courses & certifications](https://aidefensestack.com/categories/courses-certs/): 12 options. Free labs, games, courses and professional certifications for AI security, with cost and format as each provider states them.
- [All tools](https://aidefensestack.com/tools/): every tool with filters for type, deployment, OWASP risk and pricing model.
- [Open-source tools](https://aidefensestack.com/open-source/): 52 open-source projects with license and maintenance status.

## Key guides
- [AI governance platforms: what they do for EU AI Act and ISO/IEC 42001 work](https://aidefensestack.com/guides/ai-governance-platforms/): AI governance platforms compared: inventory, risk assessment, policy and evidence, plus which frameworks each vendor names, with every fact sourced.
- [Choosing an AI red teaming platform: a buyer's guide](https://aidefensestack.com/guides/ai-red-teaming-platforms/): How to choose an AI red teaming platform: when it beats open source, a 12-point buyer checklist and sourced facts on eight platforms and managed services.
- [AI security certifications and courses worth your time in 2026](https://aidefensestack.com/guides/ai-security-certifications/): AI security certifications and courses compared by role: provider, format, cost and duration as each provider states them, plus free ways to practice.
- [The best LLM guardrails in 2026, open source and commercial](https://aidefensestack.com/guides/best-llm-guardrails/): Compare LLM guardrails: open-source libraries, cloud provider APIs and hosted firewalls, with license, deployment and OWASP coverage from sourced facts.
- [GenAI data loss prevention: tools that keep sensitive data out of LLMs](https://aidefensestack.com/guides/genai-data-loss-prevention/): GenAI data loss prevention compared: tools for the user, the app and the data, from Presidio to browser DLP and vaults, with every fact sourced.
- [How to scan ML models for malware before you load them](https://aidefensestack.com/guides/scan-ml-models-for-malware/): Scan ML models for malware before you load them: why pickle files can run code, a four-step process, and scanners, signing and safe formats compared.
- [MCP security tools: how to scan and protect Model Context Protocol servers](https://aidefensestack.com/guides/mcp-security-tools/): MCP security tools compared: scanners, gateways, sandboxes and detection and response, with tool poisoning, rug pulls and token passthrough explained.
- [Open-source AI red teaming tools compared](https://aidefensestack.com/guides/open-source-ai-red-teaming-tools/): Compare open-source AI red teaming tools: scanners, attack frameworks and eval harnesses, with license, dated GitHub stars and OWASP coverage sourced.
- [Prompt injection detection tools: how they differ and where they stop](https://aidefensestack.com/guides/prompt-injection-detection-tools/): Compare prompt injection detection tools: classifier models, cloud APIs and hosted firewalls, with license, deployment and OWASP LLM01 coverage sourced.
- [Securing AI agents: a control map for tool use, identity and runtime](https://aidefensestack.com/guides/securing-ai-agents/): How to secure AI agents: a seven-control map for identity, tool vetting, gateways, sandboxes, guardrails, detection and red teaming, with sourced tools.
- [What is AI-SPM? AI security posture management explained](https://aidefensestack.com/guides/what-is-ai-spm/): What is AI-SPM? AI security posture management finds AI assets and checks them for risk. See how it differs from CSPM and DSPM and what to ask vendors.
- [All guides](https://aidefensestack.com/guides/)

## Comparisons
- [Amazon Bedrock Guardrails vs Azure AI Content Safety](https://aidefensestack.com/compare/amazon-bedrock-guardrails-vs-azure-ai-content-safety/)
- [DeepTeam vs promptfoo](https://aidefensestack.com/compare/deepteam-vs-promptfoo/)
- [garak vs PyRIT](https://aidefensestack.com/compare/garak-vs-pyrit/)
- [Lakera Guard vs Google Cloud Model Armor](https://aidefensestack.com/compare/lakera-guard-vs-google-model-armor/)
- [LLM Guard vs NeMo Guardrails: status and alternatives](https://aidefensestack.com/compare/llm-guard-vs-nemo-guardrails/)
- [ModelScan vs Fickling](https://aidefensestack.com/compare/modelscan-vs-fickling/)
- [NeMo Guardrails vs Guardrails AI](https://aidefensestack.com/compare/nemo-guardrails-vs-guardrails-ai/)
- [promptfoo vs garak](https://aidefensestack.com/compare/promptfoo-vs-garak/)
- [All comparisons](https://aidefensestack.com/compare/)

## Frameworks
- [Frameworks](https://aidefensestack.com/frameworks/): how the OWASP LLM Top 10 and MITRE ATLAS relate to the tools we list.
- [OWASP Top 10 for LLM Applications 2025, mapped to tools](https://aidefensestack.com/frameworks/owasp-llm-top-10/): 10 risk pages, LLM01 to LLM10.
- [MITRE ATLAS, mapped to AI security tools](https://aidefensestack.com/frameworks/mitre-atlas/): data version 2026.09.

## AI security layers
- [AI Security Layers](https://aidefensestack.com/frameworks/ai-security-layers/): our editorial model, not a standard. Counts: primary + also-cover.
- [Governance and risk](https://aidefensestack.com/frameworks/ai-security-layers/governance-risk/): layer 1, 7 + 0.
- [Discovery and posture](https://aidefensestack.com/frameworks/ai-security-layers/discovery-posture/): layer 2, 8 + 20.
- [Model and supply chain](https://aidefensestack.com/frameworks/ai-security-layers/supply-chain/): layer 3, 9 + 8.
- [Testing and red teaming](https://aidefensestack.com/frameworks/ai-security-layers/testing-red-teaming/): layer 4, 24 + 9.
- [Data protection](https://aidefensestack.com/frameworks/ai-security-layers/data-protection/): layer 5, 10 + 17.
- [Runtime guardrails](https://aidefensestack.com/frameworks/ai-security-layers/runtime-guardrails/): layer 6, 25 + 11.
- [Agent and tool access](https://aidefensestack.com/frameworks/ai-security-layers/agent-tool-access/): layer 7, 26 + 12.
- [Monitoring and response](https://aidefensestack.com/frameworks/ai-security-layers/monitoring-response/): layer 8, 2 + 24.

## Answers
- [AI red teaming: questions and answers](https://aidefensestack.com/answers/ai-red-teaming/): 9 questions.
- [GenAI data protection: questions and answers](https://aidefensestack.com/answers/genai-data-protection/): 9 questions.
- [LLM guardrails: questions and answers](https://aidefensestack.com/answers/llm-guardrails/): 8 questions.
- [MCP security: questions and answers](https://aidefensestack.com/answers/mcp-security/): 9 questions.
- [Model supply chain security: questions and answers](https://aidefensestack.com/answers/model-supply-chain-security/): 8 questions.
- [Prompt injection: questions and answers](https://aidefensestack.com/answers/prompt-injection/): 8 questions.
- [Superintelligence safety and AI security: questions and answers](https://aidefensestack.com/answers/superintelligence-safety/): 9 questions.
- [All answers](https://aidefensestack.com/answers/)

## Glossary
- [AI security glossary](https://aidefensestack.com/glossary/): 57 terms explained plainly, each with an anchor such as https://aidefensestack.com/glossary/#guardrails.

## Methodology: Coverage Score (rubric version 1.0)
The Coverage Score (0 to 100) measures how much a tool documents publicly. It is computed in code from sourced Facts. It does not measure detection accuracy, false positives, latency, support or price-performance, and we do not run the products. Scores compare tools within one category.

| Criterion | Points |
|---|---|
| Risk coverage | 25 |
| Capability breadth | 20 |
| Deployment flexibility | 10 |
| Openness | 10 |
| Documentation | 10 |
| Integrations | 10 |
| Maintenance | 5 |
| Pricing transparency | 5 |
| Trust transparency | 5 |

- [Methodology](https://aidefensestack.com/methodology/): sourcing standard, Fact schema and the full rubric.

## Policies
- [About](https://aidefensestack.com/about/)
- [Editorial and AI-use policy](https://aidefensestack.com/editorial-policy/)
- [How we make money (disclosure)](https://aidefensestack.com/disclosure/)
- [Corrections](https://aidefensestack.com/corrections/)
- [Advertise or get listed](https://aidefensestack.com/advertise/)
- [Privacy policy](https://aidefensestack.com/privacy/)
- [Your privacy choices](https://aidefensestack.com/privacy-choices/)
- [Cookie policy](https://aidefensestack.com/cookie-policy/)
- [Terms of use](https://aidefensestack.com/terms/)
- [Security policy](https://aidefensestack.com/security/)
- [Contact](https://aidefensestack.com/contact/)

## Optional
- [llms-full.txt](https://aidefensestack.com/llms-full.txt): every tool, guide, answer and glossary definition as plain text.
- [Sitemap](https://aidefensestack.com/sitemap-index.xml)