Illustrative sample — not a customer assessment
Example technical report structure for an AI SaaS application.
Overall AI Risk Score
74/100
Remediation Required
Testing identified one critical, one high, and one medium AI-specific security finding. The most significant risk involved insufficient authorization around tool execution. Remediation should focus on deterministic permission enforcement, confirmation boundaries, and treating retrieved content as untrusted.
The agent could be manipulated into invoking an internal workflow without sufficient confirmation.
A crafted indirect instruction caused the agent to prepare and execute a privileged tool call outside the user’s intended task.
Require server-side authorization for every tool action, add explicit confirmation for high-impact operations, and restrict tool scopes by user and task.
A malicious document in the retrieval layer influenced model behavior and displaced trusted instructions.
Injected instructions inside an indexed document were followed when the document was retrieved for a normal user query.
Treat retrieved content as untrusted data, separate instructions from context, apply retrieval access controls, and test adversarial documents continuously.
The assistant exposed internal behavioral instructions under a repeatable adversarial prompt sequence.
The response revealed internal policy fragments and routing logic that were not intended for end users.
Remove secrets from prompts, minimize sensitive operational detail, add output checks, and assume prompt contents may eventually be disclosed.
A real deliverable includes validated evidence, business impact, technical remediation, and retest recommendations for the assessed system.
Request a Free Review