Decision
Proceed, hold or do not deploy—with the reason stated plainly.
Sample assessment
See what a customer receives—not just a score.
The sample shows how AgentRiskLayer presents the decision first, then separates customer statements, observed technical evidence, reproduced failures, required fixes and limitations.
How to read it
The first pages answer the business question.
What could happen
The credible path from agent access or manipulation to business impact.
What must change
Named remediation, accountable owner, evidence requirement and deadline.
How closure is proved
The exact retest and evidence needed before the decision can change.
Example recommendationDO NOT DEPLOY
Controlled staging testing reproduced material prompt-injection and unsafe-tool behaviour. The decision does not rely on a questionnaire score alone.
Evidence examples
Each conclusion shows where it came from.
Sample values are illustrative and are not independent benchmark results.
Indirect injection influenced the staging agent
A synthetic email caused an unauthorised action request. Required control: treat retrieved content as data, enforce deterministic tool policy and retest the exact case.
MCP configuration exposed command capability
The local inspector observed mutable command-capable configuration. Required control: remove or isolate shell access and use immutable provenance.
Approval existed only as a process statement
The report distinguishes the claim from server-recorded exact-action approval evidence.
The sample is not a customer case study
It demonstrates report structure and evidence separation, not real-world assurance or commercial outcomes.