How AI agent security solutions are scored
One research plan covered every vendor on 16 Nov 2026; the sixteen evidence files here hold 902 facts, and each quote was checked against its page. One scorer scored blind, a second blind reviewer checked, and weights were committed before any evidence. The five layers weigh 6 each.
Facts read on .
The weights
The weights were fixed on , before any score existed, and have not changed since.
| Criterion | What it measures | Weight |
|---|---|---|
| Coverage: endpoint layer | Agents running on laptops and developer machines (desktop agents, coding agents, command-line agents, local MCP servers), as documented | 6 |
| Coverage: browser layer | Browser agents, AI browsers and AI extensions, as documented | 6 |
| Coverage: cloud workspace layer | Agents built or hosted in SaaS and cloud platforms (Microsoft Copilot Studio, Salesforce Agentforce, ServiceNow, AWS Bedrock agents and similar), as documented | 6 |
| Coverage: identity layer | Agent identities, credentials, tokens and permissions: inventory, least privilege, revocation, as documented | 6 |
| Coverage: gateway layer | A gateway, proxy or API control point between agents and models, tools or MCP servers, as documented | 6 |
| Stops the action | Blocking a harmful agent action before it completes, with the action types named, against detection or alerting after the fact | 20 |
| Visibility | An inventory of agents and what each can reach (tools, data, permissions) with activity logs | 15 |
| Deployment | How it deploys (agentless, API, extension, agent, proxy), what it needs, time to deploy as published | 10 |
| Policy and governance | The policy model, audit trail, and integrations with SIEM, EDR, the identity provider and ticketing | 10 |
| Transparency | Public documentation readable without sign-in, a published price, published research | 10 |
| Proof | Named customers, funding, independent analyst coverage | 5 |
Weights by situation
| Situation | Coverage: endpoint layer | Coverage: browser layer | Coverage: cloud workspace layer | Coverage: identity layer | Coverage: gateway layer | Stops the action | Visibility | Deployment | Policy and governance | Transparency | Proof |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Ranked for the endpoint layer | 30 | 0 | 0 | 0 | 0 | 20 | 15 | 10 | 10 | 10 | 5 |
| Ranked for the browser layer | 0 | 30 | 0 | 0 | 0 | 20 | 15 | 10 | 10 | 10 | 5 |
| Ranked for the cloud workspace layer | 0 | 0 | 30 | 0 | 0 | 20 | 15 | 10 | 10 | 10 | 5 |
| Ranked for the identity layer | 0 | 0 | 0 | 30 | 0 | 20 | 15 | 10 | 10 | 10 | 5 |
| Ranked for the gateway layer | 0 | 0 | 0 | 0 | 30 | 20 | 15 | 10 | 10 | 10 | 5 |
Score anchors
- 0: none found on the read date
- 1-3: mentioned, partial, or limited to one surface, one layer or one action type
- 4-6: documented and working, with gaps the documentation shows
- 7: the most a vendor claim or vendor documentation alone can earn: detailed, specific, covering the whole criterion
- 8-10: detailed and confirmed by a second source (third-party write-up, or observed directly in public documentation, code, a marketplace listing or a release record)
Rules
- Every vendor is scored 0 to 10 on every criterion, whole numbers, from the evidence file only, by a scorer that works from coded packets with every vendor name removed.
- A vendor claim scores at most 7 until a second source confirms it (kind third-party or observed).
- None found scores 0 and is written "none found on the read date".
- Breadth facts (module counts, integration counts, headcount, funding) inform Proof only, which weighs 5, so they carry at most 10 percent of any ranking.
- The total is the weighted sum of the scores divided by 10 (0 to 100 scale, published to one decimal).
- Equal totals print as a joint rank (for example joint 2nd); inside a joint rank vendors are listed alphabetically for display only.
- Once the first score exists, no weight and no criterion changes. A score changes only when a sourced fact changes, logged with the fact.
How are AI agent security solutions scored?
Each vendor is scored 0 to 10, in whole numbers, on eleven criteria, from the evidence file only. The total is the weighted sum divided by 10, giving a 0 to 100 scale printed to one decimal.
The same research plan ran for every vendor on 16 Nov 2026: vendor pages, documentation, pricing pages and release notes, plus at least two third-party hosts per vendor. It produced 1,469 facts across 32 vendor files. Every fact carries a source URL, a short verbatim quote and a read date, and each quote was checked against its page.
Why do the five layers weigh 6 each?
The five layer criteria (endpoint, browser, cloud workspace, identity and gateway) weigh 6 each, 30 in all, so no single place agents run can dominate the site ranking. That equal split is a choice made before scoring, and it is recorded in the project decisions log.
Each layer view moves all 30 points onto its own layer and sets the other four layers to 0. The other six criteria keep their weights: stopping the action 20, visibility 15, deployment 10, policy 10, transparency 10 and proof 5.
How does blind scoring work?
One scorer reads the evidence file for each vendor with each vendor's own name, parent and hosts removed from its own record and replaced by a code, and scores every criterion with a short written reason and the fact ids behind it. A second reviewer, also blind, checks the scores against the same facts and flags any that the evidence does not support.
Scores change only when a sourced fact changes, and the change is logged with the fact.
What do the score anchors mean?
A 0 means none found on the read date, 1 to 3 means mentioned or limited to one surface, one layer or one action type, and 4 to 6 means documented and working with gaps the documentation shows. A 7 is the most a vendor claim or vendor documentation alone can earn.
A score of 8 to 10 needs detail confirmed by a second source: a third-party write-up, or direct observation of public documentation, code, a marketplace listing or a release record.
How are corrections handled?
A vendor or reader can send a sourced correction to the address on the corrections page. If the source changes a fact, the fact, any score that rests on it and the dated changelog entry are updated together.
Weights and criteria stay fixed once the first score exists, so a correction can move a score but not the rules.
Changelog
- : weights and criteria fixed.
- : evidence read for every vendor by one research plan.
- : scores published (blind: one scorer, then one blind reviewer).
- : this site built from the committed scores.