Method
Finder rules: how the GenAI security quiz matches vendors
In short
We read each vendor's public pages, record what they describe for three control approaches and thirteen practical checks, and apply published rules to your quiz answers. There are no scores or rankings. The same rules apply to every vendor, and anything we did not find is marked Not in our sources rather than guessed.
By The Inline Desk · GenAI Security Finder · Published 2026-10-26
Desk research from the six vendors' own public pages, read on 7 October 2026. Profiles follow our published method; we did not run the products.
What do we record for each vendor?
For each of the three approaches (AI controls in a SASE or SSE cloud, a standalone AI security layer, a firewall or self-hosted AI gateway) we record one of three statuses with a note and a source: Described, Partly described (a narrower scope, or the enforcement point is not stated) and Not in our sources. We also record thirteen checks, each Stated, Partly stated or Not in our sources: Finds shadow AI use, Inspects prompts for sensitive data, Controls AI agent and MCP traffic, Covers coding agents, Protects homegrown AI apps at runtime, Blocks inline in real time, Coaches or warns the user, Has a watch or log-only action, Redacts or anonymizes data in prompts, Can run on your own infrastructure, States FedRAMP, Lists security attestations and One console for AI and network policy. Every status carries the vendor page it came from and the date we read it.
How does the quiz weigh the approaches?
| Question | Answer | Points |
|---|---|---|
| Q1 What worries you most about generative AI at work? | People using AI tools we have not approved (shadow AI) | SSE +3 |
| Q1 What worries you most about generative AI at work? | Sensitive data pasted or uploaded into prompts | SSE +2, AIG +1 |
| Q1 What worries you most about generative AI at work? | AI agents and MCP servers acting on our systems | SSE +1, AIG +3 |
| Q1 What worries you most about generative AI at work? | Coding agents such as Claude Code, Cursor or Codex | SSE +2, AIG +2 |
| Q1 What worries you most about generative AI at work? | AI apps we build ourselves (prompt injection, jailbreaks) | AIG +4, FWG +1 |
| Q2 When a risky AI interaction is found, what should happen first? | Block it inline, in real time | SSE +2, FWG +1 |
| Q2 When a risky AI interaction is found, what should happen first? | Remove the sensitive data and let the rest through | SSE +1, AIG +1 |
| Q2 When a risky AI interaction is found, what should happen first? | Warn the person and let them decide | SSE +2 |
| Q2 When a risky AI interaction is found, what should happen first? | Only watch and log for now | SSE +1 |
| Q3 What network security carries your users' traffic today? | A SASE or SSE platform | SSE +4 |
| Q3 What network security carries your users' traffic today? | Firewalls at our sites, no SASE or SSE yet | SSE +1, FWG +4 |
| Q3 What network security carries your users' traffic today? | Neither: we are choosing now | SSE +2, AIG +1 |
| Q5 Where may AI traffic be inspected? | In the vendor's cloud is fine | SSE +2, AIG +1 |
| Q5 Where may AI traffic be inspected? | It must stay on our own infrastructure | FWG +4 |
| Q5 Where may AI traffic be inspected? | No preference | No points |
| Q6 Which compliance requirement applies to the vendor? | FedRAMP (US federal) | No points |
| Q6 Which compliance requirement applies to the vendor? | Standard attestations such as SOC 2 or ISO 27001 | No points |
| Q6 Which compliance requirement applies to the vendor? | None yet | No points |
| Q7 How many people use AI tools for work? | Under 1,000 | SSE +1 |
| Q7 How many people use AI tools for work? | 1,000 to 10,000 | No points |
| Q7 How many people use AI tools for work? | Over 10,000, or many sites and countries | SSE +1, AIG +1 |
Approaches are ordered by points, highest first; ties keep the fixed order SSE, AIG, FWG. The first is Start here. Each other approach is Also relevant if its points are at least the larger of 3 and half the first approach's points. The rest are Lower on your list (points above 0) or Not suggested by your answers (0 points). An empty answer to the first question counts as shadow AI.
How does the quiz check vendors?
Each vendor gets 2 points for every picked approach (Start here and Also relevant) its pages describe, 1 for partly described and 0 for not in our sources. Then each check your answers trigger adds points as below.
| Check | When it applies | Stated | Partly | Not in our sources |
|---|---|---|---|---|
| Finds shadow AI use | You worry about shadow AI | 2 | 1 | 0 |
| Inspects prompts for sensitive data | You worry about sensitive data in prompts | 2 | 1 | 0 |
| Controls AI agent and MCP traffic | You worry about AI agents and MCP | 2 | 1 | 0 |
| Covers coding agents | You worry about coding agents | 2 | 1 | 0 |
| Protects homegrown AI apps at runtime | You worry about AI apps you build | 2 | 1 | 0 |
| Blocks inline in real time | Your first response is to block inline | 2 | 1 | 0 |
| Redacts or anonymizes data in prompts | Your first response is to remove the sensitive data | 2 | 1 | 0 |
| Coaches or warns the user | Your first response is to warn the person | 2 | 1 | 0 |
| Has a watch or log-only action | Your first response is to only watch | 2 | 1 | 0 |
| Can run on your own infrastructure | Traffic must stay on your own infrastructure | 2 | 1 | 0 |
| States FedRAMP | You need FedRAMP | 2 | 1 | 0 |
| Lists security attestations | You need standard attestations | 2 | 1 | 0 |
| One console for AI and network policy | Over 10,000 people or many sites | 2 | 1 | 0 |
| Extends a platform you already run | The vendor you picked in Q4 | +3 | ||
Vendors with the same total are shown as equals and listed alphabetically. We never show the total as a score. Pricing is not a check because none of the six vendors publishes prices.
What does Not in our sources mean, and is it fair?
It means we did not find the item on the pages we read. It counts zero, not against the vendor. It does affect order: a vendor whose pages state a redaction action can meet a redaction check, and one whose pages do not cannot. Large vendors spread facts across many pages, and we read the main product and AI security pages for each. If a vendor offers more than its pages say, the fix is a source, not a guess: send it through the contact page and we will add it.
Why these six vendors?
Cato Networks, Cisco, Fortinet, Netskope, Palo Alto Networks and Zscaler. Each sells network security from the cloud (SASE or SSE) and also publishes AI security capabilities for employee AI use, agents or homegrown AI apps. This edition covers these six only.
Where do the facts come from?
- Vendor product pages, trust and compliance pages and newsroom items, all read on 7 October 2026. Each fact carries its source link and read date.
- Vendor figures such as app counts, PoP counts, uptime SLAs and attack-scenario counts are quoted as the vendor's statement and are never our finding. Network figures are stated on different bases, so we show each vendor's own wording and never compare them.
- Analyst placements are quoted only as each vendor states them on its own page, with the report date where the page gives it.
- Where a vendor does not publish a price, we write Not published. None of the six publishes list prices.
What are the limitations?
Public sources only. No hands-on testing, no demos, no vendor interviews. A vendor's pages describe what it sells, not how well it works in your environment. AI security products change quickly and several of these capabilities are new. Use the finder to decide what to ask and what to put into a pilot, not as a test result.
How do we handle corrections?
Send a correction through the contact page with the topic Correction and a public source. We check it against the vendor's page and update the data; quiz results recompute automatically.
Questions
Did you test these products?
No. Every fact comes from public vendor pages read on 7 October 2026.
Can a vendor pay to appear first?
No. Order comes only from the published rules and your answers, and equal fits are shown alphabetically.
Why no scores?
Because the right control depends on your situation more than on any single number. A score would hide the constraint that matters to you, such as FedRAMP, a SASE platform you already run, or a need to keep AI traffic on your own infrastructure.