About AI Defense Stack

What this directory is

AI Defense Stack is an independent directory of tools that protect AI systems: LLM firewalls and guardrails, AI red teaming and evaluation, model and ML supply-chain security, AI security posture management, agent and MCP security, GenAI data-leak prevention, AI governance, and training courses. Each listing is a short page with the same fields, so two tools can be compared without reading two marketing sites.

We compile from public primary sources: the vendor's own documentation, pricing and product pages, the project's repository and license file, and official standards pages such as OWASP and MITRE. We do not run or benchmark the tools ourselves, and we say so wherever a score appears.

Who writes it

The directory is written and maintained by the AI Defense Stack team. We publish under the team name, not under individual names, and we do not invent personas or claim experience we cannot show. Text is drafted with AI assistance and every factual statement is checked against the primary source cited beside it. The full process is in our editorial and AI-use policy.

What makes it different

• Verified facts

  Each fact carries an evidence chip with its source host and the date we checked it. A fact we cannot verify is left out, not guessed.

• A transparent score

  The Coverage Score is computed from sourced attributes by a published formula. Nobody types a score, and sponsorship cannot change it. See the methodology.

• Open source and commercial, side by side

  Filters work everywhere. Open-source tools show their license, repository, GitHub stars and maintenance status as of a date.

• Framework mapping

  Tools are mapped to the OWASP Top 10 for LLM Applications and, where it fits, MITRE ATLAS. The mapping is labeled "Our analysis" and tied to a sourced capability.

• Honest status

  Archived and acquired products are labeled. Comparisons say where each side is stronger.

• Machine-readable

  We publish llms.txt and clean structured data so AI assistants and search engines can cite correct facts.

How we stay independent

We may earn money from affiliate links, sponsored listings and ads. What is live today, and the rules for each, are on our disclosure page. In short: commercial relationships are always labeled, they never change a Coverage Score, a ranking order, a pick or a comparison verdict, and we can decline any sponsor. A tool page says plainly when we have no commercial relationship with the vendor.

Vendors can ask us to correct a fact through the corrections page. We fix errors we can confirm at a primary source, and we do not take payment to change what we publish.

What we do not do

• We do not publish star-style scores, user counts or testimonials, and we make no claims about who uses a tool.
• We do not show vendor logos or product screenshots. Tool icons are generated monogram tiles and our illustrations are original.
• We do not give legal, compliance or purchasing advice. Check the vendor before you buy, and ask your own advisers about regulation.
• We do not claim a tool makes you compliant with any law. We report what the vendor states it supports, with the source.

Who it is for

• Security engineers who need to pick and wire up a control this quarter, such as a guardrail, a scanner or red-team tooling.
• AppSec and product security leads who own LLM features and need a defensible shortlist.
• CISOs and security architects weighing GenAI controls, who need to know what each category covers and what it does not.
• ML engineers and MLOps teams who own the model supply chain and evaluation pipelines.

Where to start

Browse the eight categories, filter all tools, or see the open-source view. If you want to know how a number or label was produced, read the methodology. To report a mistake or ask a question, write to [email protected] or see the contact page.