BastionLLM

The Free LLM Security Scanner That Finds Gaps Before Attackers Do

Paste the chat API URL your app POSTs JSON to — not the website people visit — and the request JSON it expects. We replace {{PROMPT}} with each attack and return a report your team can act on.

6 attacks with guest scan25 attacks with free plan489 test payloads in full scanOWASP LLM Top 10 coverage

The HTTPS URL your app POSTs chat JSON to — not the website customers open in a browser.

API format

Pick once. We fill the request JSON and the reply path. Edit either if your endpoint differs.

Choose a format to continue

Must match the JSON your endpoint expects. We replace {{PROMPT}} with each attack.

Choose OpenAI, Anthropic, or Custom JSON so the request matches your API.

How does the reply arrive?

This only changes how we read the response. It does not change the request format.

Dot path to the text field in the JSON response. Filled by the format above — edit it if yours is different.

Pick a format above, or type the field that holds the model text.

Choose a request format or paste JSON with {{PROMPT}} to enable the scan.

What BastionLLM does

BastionLLM is a free LLM security scanner that checks your endpoints for real attack paths, including jailbreaks, prompt injections, and data leak patterns.

You provide the endpoint details once. We run OWASP-aligned tests and return results your team can triage, fix, and verify.

Why this matters

Why this matters:

LLM applications face attack vectors that traditional security tools miss. Here's what's at stake:

  • Attackers can steer model behavior with plain text in user input, docs, or tool output.
  • Prompt injection can override rules, and jailbreaks can bypass refusal and safety logic.
  • Prompt leak attempts can expose hidden instructions, business logic, and internal context.
  • One-time pre-launch tests are not enough after model, prompt, or integration changes.
  • Repeatable OWASP LLM Top 10-aligned scans catch regressions before they reach production users.

Want the full breakdown? Read the OWASP LLM Top 10 guide.

Here's how it works

Here's how it works:

Connect your endpoint once. BastionLLM handles the probing, scoring, and reporting.

  1. Connect your real endpoint with its request JSON, headers, and response parsing paths.
  2. We send controlled adversarial prompts across prompt injection, jailbreak, and leak tests.
  3. We score each payload and show raw outputs so engineers can reproduce each failure.
  4. Run scheduled scans to detect regressions as prompts, models, and integrations evolve.

See how this maps to each OWASP category in the OWASP LLM Top 10 guide.

What you get

What you get:

Jailbreak Testing

  • What: Simulated DAN and role-play bypass prompts.
  • Why: Finds weak refusal logic before release.
  • How: Score outputs by policy break severity.

Prompt Injection

System Prompt Leaks

  • What: Prompt extraction and disclosure probes.
  • Why: Prevents leakage of hidden instructions.
  • How: Detect direct leaks and partial prompt hints.

Get the full OWASP category map in the OWASP LLM Top 10 guide.

Blog

New from the BastionLLM blog

Deep dives on prompt injection, jailbreaks, and practical testing workflows for production LLM systems.

Start with the OWASP LLM Top 10 guide for the full framework.

View all articles

Featured article

5 Prompt Injection Attacks Your LLM Endpoint Isn't Ready For in 2026

Five high-success attack patterns that still bypass production defenses, plus how to test your endpoint before attackers do.

Frequently Asked Questions

What is BastionLLM?

BastionLLM is a free LLM security scanner. It tests AI endpoints for prompt injection (where attackers override model instructions through crafted input), jailbreaks (prompts that bypass safety rules), and system prompt leaks (exposing hidden instructions). Tests map to the OWASP LLM Top 10.

What is prompt injection in LLMs?

Prompt injection is an attack where malicious input manipulates an LLM into ignoring its instructions or leaking sensitive data. BastionLLM tests your endpoints against known prompt injection patterns.

Is BastionLLM free to use?

Yes, BastionLLM offers a free tier that allows you to scan LLM endpoints for common security vulnerabilities including jailbreaks and system prompt leaks.

How to test an LLM for jailbreaks?

You can test your LLM for jailbreaks by sending adversarial prompts designed to bypass safety filters. BastionLLM automates this by running multiple jailbreak patterns (like DAN) and scoring the model's response.

What is the OWASP LLM Top 10?

The OWASP LLM Top 10 is a list of the most critical security vulnerabilities for Large Language Model applications. BastionLLM maps its security probes directly to these risks, including prompt injection (LLM01) and sensitive information disclosure (LLM06).

What do I need before running a scan?

You need the chat API URL your app POSTs JSON to (not the website people visit), the request JSON with {{PROMPT}} in the user-message field, and the response field path. Add any auth headers your endpoint requires.

Can I test staging or internal pre-production environments?

Yes. If the endpoint is reachable from BastionLLM and you are authorized, you can test staging or pre-production safely before release.

Is one scan enough for LLM security?

Usually no. Model behavior changes after prompt edits, model upgrades, and integration updates, so repeated scans catch regressions early.

Still have questions? The OWASP LLM Top 10 guide covers every category in depth.