Why verdicts matter
Every product in our rankings documents blocking. If blocking were enough, the category would be simple. In practice, a rule strict enough to stop a dangerous command will also stop some legitimate ones, and a developer who loses an hour to a false block will look for a way around the control. The other verdict types exist to keep work moving while still changing the outcome.
The six types we track
- Block or deny. Stop the action before it runs. Every vendor we score documents this.
- Ask a person. Pause the action and ask the user, or an approver, whether it should run. Bay describes an Ask verdict alongside Allow and Deny. Onyx Security lists ask as one of its five actions.
- Alert or warn. Let the action run while warning the user or alerting the security team. Harmonic Security describes warning the employee with context or logging silently. Onyx Security lists alert, and Lasso Security's MCP gateway can alert.
- Mask or modify. Change the action or its data before it continues. Onyx Security lists mask and steer. Zenity describes modifying a dangerous tool call. Lasso Security describes sanitizing MCP traffic and masking sensitive data before prompts reach third-party models.
- Simulate first. Run a rule in a mode that records what it would have decided before it enforces anything. Bay describes a Simulation Mode for new rules.
- Remove or scale back. Remove a risky tool, extension or package, or cut its permissions. Bay describes MCP Control that removes disallowed MCP servers in Enforcement mode. Koi describes one-click removal of insecure packages, extensions and AI tools. Bloom Security describes removing risky tools, revoking permissions and scaling back overpermissioned MCP servers.
How we score it
The verdict range criterion is computed, not judged: the number of verdict types a vendor documents, divided by six, times ten, rounded. On the pages we read, Bay and Onyx Security document four each, the most of the scored products, but different ones: Bay documents ask, simulate and remove, Onyx Security ask, alert and mask. The rankings page shows the full census, with the vendor's own words and a source for each row.
When each verdict helps
Ask is the verdict for actions that are sometimes right and sometimes not, such as a force push or a package install from a new registry: the person who started the agent usually knows. Warn suits policies you are still learning, where you want people to see the rule before it bites. Mask and modify suit data rules: let the prompt go, without the customer record in it. Simulate suits every new rule, because it shows the false positives before anyone is blocked. Remove suits inventory findings: an extension nobody approved, an MCP server with more permissions than it needs.
What the census does not tell you
A documented verdict is a description on a vendor page. It does not tell you how the verdict feels to the user, how often it fires or whether it applies to every agent. Use the census to build a shortlist, then ask each vendor to demonstrate the verdicts you need on your own agents.