How to Send Your Manager an LLM Security PDF Without a Credit Card

Why a screenshot is not enough

Most developers who want LLM security testing hit the same wall. They will not ask a manager for a company card until they have something to show. A chat screenshot of one jailbreak is easy to dismiss. A PDF with pass/fail verdicts, the payloads you sent, and the model replies is something a manager can forward.

You do not need Pro for that first artifact. Guest scans need no account. A free BastionLLM account needs no credit card. You scan a chat API — usually staging — and export a PDF. Pro is for weekly retests and the 450-plus attack pack after you already have evidence.

What you need

You need the URL your application POSTs chat JSON to, not the website customers open. If people visit https://app.example.com and your frontend calls https://api.example.com/v1/chat, scan the second URL.

• Authorization to test it. Use staging or a dedicated test index when you can.
• Request JSON with {{PROMPT}}. BastionLLM replaces that placeholder with each attack.
• Where the model text lives in the JSON reply, for example response.answer or choices.0.message.content.
• Headers if the API requires them, such as Authorization.

If you are still mapping the request shape, the example endpoint shows a working payload. The no-setup prompt injection walkthrough covers guest scans in more detail.

Scan the vulnerable demo first

If you have never used the scanner, run the built-in demo before you point it at your company API. On bastionllm.com click Scan the vulnerable demo. No URL, headers, or schema. The demo endpoint is designed to fail so you can see a red report.

After you create a free account, the empty dashboard offers the same one-click demo. That path saves the endpoint, runs 25 attacks, and lands you on a report you can export. Use it to learn the UI. Do not send the demo PDF as proof that your product is safe or broken.

Scan your staging chat API

Create a free account at /register. No card. On the dashboard choose Add your endpoint.

1. Paste the staging chat API URL (HTTPS POST).
2. Pick OpenAI-compatible, Anthropic, or custom JSON. Edit the body so it matches what your app sends, with {{PROMPT}} where the user message goes.
3. Set the response field path to the text field in the reply.
4. Add headers if staging requires an API key or bearer token.
5. Confirm you own the endpoint, then run the scan.

Free runs 25 attacks per scan on one saved endpoint. That is enough for a first evaluation. You do not need to verify ownership to get a PDF. Verification unlocks Pro coverage later, not the report you are about to send.

Export the PDF and share it

When the scan finishes, open the report. You will see totals, SAFE / VULNERABLE verdicts, the prompt sent, and the model response. Click Export PDF. JSON is there if an engineer wants the raw dump.

You can also copy the public report link from the Share control. That URL does not include raw model replies, which is what you want if the link might be forwarded widely. The PDF you download while logged in includes the responses, which is what a manager needs to see that this is not a theoretical finding.

If a verdict looks wrong, read the response before you escalate. The scan results guide covers false positives and how to prioritize fixes.

What to put in the email

Keep it short. Attach the PDF. Do not ask for a card in the first mail unless you already know you need weekly coverage.

When you actually need Pro

Stay on Free until you have a report and a reason to retest. Upgrade when you need more than a one-off:

• Three saved endpoints instead of one
• 450-plus attacks per scan on a verified endpoint
• Weekly automated scans
• 12-month history

Pro is €49/month or €449/year, tax included. The 30-day trial requires a card at checkout; you are billed when the trial ends. Pricing is on /pricing.

FAQs

Streaming setup is covered in how to configure SSE and NDJSON endpoints.