AI Agent Security Solutions

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.

Criteria and weights (out of 100)
CriterionWhat it measuresWeight
Coverage: endpoint layerAgents running on laptops and developer machines (desktop agents, coding agents, command-line agents, local MCP servers), as documented6
Coverage: browser layerBrowser agents, AI browsers and AI extensions, as documented6
Coverage: cloud workspace layerAgents built or hosted in SaaS and cloud platforms (Microsoft Copilot Studio, Salesforce Agentforce, ServiceNow, AWS Bedrock agents and similar), as documented6
Coverage: identity layerAgent identities, credentials, tokens and permissions: inventory, least privilege, revocation, as documented6
Coverage: gateway layerA gateway, proxy or API control point between agents and models, tools or MCP servers, as documented6
Stops the actionBlocking a harmful agent action before it completes, with the action types named, against detection or alerting after the fact20
VisibilityAn inventory of agents and what each can reach (tools, data, permissions) with activity logs15
DeploymentHow it deploys (agentless, API, extension, agent, proxy), what it needs, time to deploy as published10
Policy and governanceThe policy model, audit trail, and integrations with SIEM, EDR, the identity provider and ticketing10
TransparencyPublic documentation readable without sign-in, a published price, published research10
ProofNamed customers, funding, independent analyst coverage5

Weights by situation

Weights used for each situation view
SituationCoverage: endpoint layerCoverage: browser layerCoverage: cloud workspace layerCoverage: identity layerCoverage: gateway layerStops the actionVisibilityDeploymentPolicy and governanceTransparencyProof
Ranked for the endpoint layer30000020151010105
Ranked for the browser layer03000020151010105
Ranked for the cloud workspace layer00300020151010105
Ranked for the identity layer00030020151010105
Ranked for the gateway layer00003020151010105

Score anchors

Rules

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