Our 18-category taxonomy tells you what an AI tool does. The security risk layer tells you what it puts at risk — classifying AI destinations by data sovereignty, abusive purpose, and whether they train on the data your users submit. Because a model is not a domain, we classify what can actually be enforced at the firewall and DNS layer: the destination.
“AI tool” is not a risk verdict — Canva is an AI tool, and so is an uncensored chatbot that trains on every prompt. Functional categories answer policy questions like “should designers use image generators?”. Risk categories answer security questions like “which of the 20,183+ AI destinations on our network sends data somewhere we cannot accept?”. You need both axes to write a defensible AI policy.
Text & Language, Image & Visual, Code & Development, Agents & Automation, and 14 more — with subcategories for granular acceptable-use policies. Included with every plan. Browse the taxonomy →
An overlay on the same domain inventory, flagging destinations by data-sovereignty exposure, abusive purpose, and train-on-your-data terms — so you can block by risk posture without blocking your users’ sanctioned tools.
AI services whose ownership, hosting, or legal jurisdiction places submitted data under foreign data-access laws that many organizations — and a growing number of regulators and state bans — cannot accept.
Destinations whose advertised purpose is abuse: deepfake and “nudify” generators, uncensored / jailbroken chatbots, voice-cloning for impersonation, and AI tooling marketed for phishing and fraud.
AI porn generators, adult chatbots and explicit “AI companion” platforms built for fictional adult content. Distinct from abusive-purpose AI — no real person is targeted — but unambiguously unsuitable for schools and most workplaces, and one of the fastest-growing corners of the AI web.
Tools whose consumer or free tiers use submitted content for model training under their own Terms of Service — the classic shadow-AI leak channel: an employee pastes source code or client records into a free chatbot, and it becomes training data.
A model is not observable from outside a website — but a service’s legal terms, ownership, hosting, and advertised purpose are. Our risk layer is built from those observable signals, across the same 20,183+ domain inventory that powers the functional taxonomy.
Automated collection of Terms of Service, privacy policies, and data-processing addenda for every AI destination in the inventory.
Large-language-model extraction of the clauses that matter: training-use rights, retention, data transfers, jurisdiction, and opt-out mechanics — normalized into structured risk attributes.
Corporate registration, parent-company jurisdiction, and hosting-infrastructure signals combined into the data-sovereignty assessment.
High-impact flags are human-reviewed before release, and the pipeline re-checks terms on an ongoing basis — a vendor’s ToS change can move a domain in or out of a risk category.
Each risk category is delivered as a separate feed cut, in the same hosted-endpoint formats as the main blocklist — point your firewall or DNS filter at a stable URL and updates flow automatically.
All three risk categories are delivered as an add-on to the Professional plan and included in OEM & Enterprise licensing, in every hosted-endpoint format above. MSPs: multi-tenant terms are available — tell us your fleet size.
Tell us which risk categories you need (sovereignty, abusive-purpose, trains-on-data), your filtering stack (firewall, DNS filter, proxy), and how many sites or endpoints you cover.