Offer Checklist Sample report Intake

Free diagnostic

AI Agent Risk Self-Check

Score one agentic workflow before it touches browsers, tools, files, customer records or external systems.

Risk signals

10 checks

High-impact tools The agent can send messages, update records, trigger workflows or call payment/order APIs. Shared browser session Browser automation reuses a personal or production session. Secrets in reach Secrets, tokens or unrestricted filesystem paths are reachable by the runtime. No allowlist There is no strict allowlist for domains, tools, file paths or commands. No approval gate Irreversible actions do not require human approval. Verbose sensitive logs Logs contain prompt text, tool output or customer data without redaction. No injection test No prompt-injection tests have been run against pages, files or user input. Untrusted content The agent reads untrusted web pages, PDFs, tickets or emails before choosing actions. No budget limits There are no spend caps, rate limits or retry limits. No recovery plan There is no rollback, token revocation or incident checklist.

Current score

0 / 21

Low visible risk

No high-risk signals selected yet. Keep the agent scoped, logged and approval-gated before launch.

Best next action

Use the free checklist first

The paid report should save a risky launch, customer answer or engineering decision. If this is only a toy demo, do not buy yet.

There is a real pilot, customer demo or production launch.
The workflow can act on systems, records, browsers or customer data.

Paid review is usually worth it if the agent can take actions, touch customer data or is close to launch.

View sample deliverable