Red Teaming - LockLLM Documentation

Link to section: What is Red Teaming?What is Red Teaming?

Beta, and free while it lasts. Red Teaming is in beta and assessments cost nothing at the moment. There is no charge per assessment and no credit deduction for running one. Pricing will be announced before it changes, and anything you have already run stays available to you.

Red Teaming lets you run automated security assessments against a domain you own. You verify that you control the domain, choose what you want tested, and LockLLM red teams it the way a real attacker would, then hands you a graded report with reproducible findings and clear fixes.

It is AI red teaming, extended to the whole domain. Most tools stop at the AI endpoint. Red Teaming tests your web application and API alongside the AI and LLM features they expose, so you see your real attack surface in one report.

Key benefits:

Verified ownership - assessments only run against domains you prove you control
Web and AI coverage - application security plus prompt injection, jailbreaks, and other AI-specific attacks
Assessment templates - focus a run on your whole domain, its web layer, or a specific AI endpoint
Graded reports - an overall grade, per-category results, and findings with severity and remediation
Non-destructive - issues are proven without causing damage, at a pace that stays gentle on your site
Scoped and consent-gated - you set exclusions, and ownership is re-checked right before every run

Link to section: How Assessments Are RunHow Assessments Are Run

Assessments are run by our own proprietary AI, working alongside cybersecurity experts who review what it finds.

The AI does the breadth: it explores the target, generates and adapts attacks, and works through far more of the attack surface than a manual review covers in the same time. The experts do the judgement: they check findings before they reach you, discard the ones that do not hold up, and confirm that a reported issue is genuinely reproducible.

That is why a report tells you what was proven rather than what was suspected. Findings arrive with reproduction steps and a fix, and a finding that could not be reproduced does not become a finding.

Link to section: Assessment TemplatesAssessment Templates

When you add a target, you choose a template that scopes what the assessment focuses on.

Template	Best for	What it focuses on
Full assessment	Most domains	Web application and API security plus the AI and LLM defenses on any AI features the domain exposes
Web domain	Sites and APIs without AI features	Broad web testing across the domain: injection, access control, request forgery, misconfiguration, and data exposure
AI / LLM endpoint	A specific AI or chat endpoint	Focused AI testing: prompt injection, jailbreaks, system prompt extraction, unsafe tool use, retrieval poisoning, and unsafe output

You can register the same domain more than once with different templates and endpoints - for example, a Full assessment of the whole site and a focused AI / LLM endpoint assessment of your chatbot API.

Link to section: Verifying OwnershipVerifying Ownership

Before you can run an assessment, you prove you control the domain. Choose one of two methods when you add a target.

Link to section: DNS recordDNS record

Add a TXT record to your domain with the value shown on the target, then select Verify. DNS changes can take a few minutes to appear.

Because control of your DNS proves control of the whole zone, DNS verification lets you point an assessment at the domain or any subdomain of it.

Link to section: File uploadFile upload

Host a file at https://your-domain/.well-known/lockllm-verify.txt containing the value shown on the target, then select Verify.

File verification proves control of that exact host, so the assessment must target the exact domain you verified. To cover subdomains, use DNS verification.

Verification tokens expire after a short window. If a token expires before you use it, a fresh one is issued automatically the next time you try to verify.

Link to section: Editing a TargetEditing a Target

A target is not fixed at the moment you add it. Open it and select Edit scope to change:

Label - the name the target is listed under
Scope notes - anything the assessment should know before testing, such as which features sit behind a login
Excluded paths - one path per line; anything listed here is never touched

Changes apply to assessments you start after saving.

The domain, the endpoint and the assessment template stay as you set them, because the domain is what you verified. To test a different domain or a different template, add it as a new target. In an organization, editing a target requires admin access.

Link to section: Assessment ScopeAssessment Scope

What is tested depends on the assessment template you chose. What is off limits, and the rules every assessment follows, are the same no matter which template you pick. The Red Teaming page shows the same summary, with the in-scope list switching as you select an assessment type.

Link to section: In scopeIn scope

Web application and API - covered by Full assessment and Web domain

Injection (SQL, command, template, SSRF)
Cross-site scripting and request forgery
Authentication and access control (IDOR, privilege escalation)
Security misconfiguration and exposed data
Known vulnerable components

AI and LLM features - covered by Full assessment and AI / LLM endpoint

Prompt injection (direct and indirect)
Jailbreaks and safety bypass
System prompt and data extraction
Tool and function-call abuse
Unsafe output handling and data exfiltration

If you want both areas covered in a single run, choose Full assessment.

Link to section: Out of scopeOut of scope

Any domain you have not verified
Anything the assessment template you chose does not cover
Destructive actions - data deletion, account takeover, denial of service
Physical, social engineering, or phishing attacks
Third-party services and infrastructure you do not control
Any paths you list as excluded

Link to section: Rules of engagementRules of engagement

You must own, or be authorized to test, every target
Ownership is re-verified right before each assessment
Testing is non-destructive - issues are proven without causing damage
Traffic is paced to stay gentle on your site, and you can ask for a slower pace in the scope and rules
Your users' personal data is never stored as evidence

Link to section: Running an AssessmentRunning an Assessment

Add a target. Go to Red Teaming and select Add target. Choose an assessment template, enter your domain, and optionally point to a specific endpoint. Leave the endpoint blank to assess the entire domain.
Set the scope. Write anything the assessment should know before testing - a slower pace, a feature owned by a supplier, a quiet window - and list any paths to exclude.
Verify ownership. Follow the DNS or file instructions on the target, then select Verify.
Authorize and start. Select Run assessment, confirm you are authorized to test the target, and start the run. Ownership is re-checked at this point.
Watch progress. The assessment runs through reconnaissance, testing, and reporting. You can leave the page and come back - progress is saved.
Get your report. When the assessment finishes, the report appears on the target's assessment card, and we email you so you do not have to watch for it.

Only one assessment runs against a target at a time. To run a different template or endpoint at the same time, add it as a separate target.

Link to section: Reading Your ReportReading Your Report

Each finished assessment shows:

An overall grade from A to F, with a score out of 100
Per-category results showing how many attacks were refused in each area
Findings with a severity label and remediation guidance
A downloadable PDF report with the full details

Use the grade and category breakdown for a quick read on your posture, and the findings list to prioritize fixes. Re-run an assessment after remediation to confirm the issues are resolved.

Link to section: The PDF reportThe PDF report

Select Download report on a finished assessment. The PDF contains everything on the card plus the detail that does not fit there:

Evidence for each finding - what was sent to your application, and how it responded
Recommended fixes written for that specific finding
How LockLLM would have stopped this - the gateway controls that address what this assessment actually found, with the exact header to send for each

That last section is drawn from the categories where attacks got through, not from a list of everything the platform offers. An assessment that refused everything recommends nothing.

Reports are private to the account that ran the assessment. In an organization, every member can download the organization's reports; nobody outside it can.

Link to section: Organization SupportOrganization Support

Red Teaming works in both personal and organization contexts. In an organization, targets and assessments are shared with your team. Organization admins can add targets and run assessments; all members can view targets and reports.

Link to section: FAQFAQ

Link to section: Do I need an AI endpoint to use Red Teaming?Do I need an AI endpoint to use Red Teaming?

No. The endpoint is optional. Leave it blank to assess your entire domain, choose the Web domain template for a site without AI features, or point the AI / LLM endpoint template at a specific AI endpoint.

Link to section: Is the testing safe to run against production?Is the testing safe to run against production?

Yes. Testing is non-destructive - issues are proven without causing damage - and traffic is paced to stay gentle on your site. If a target needs a slower pace or a quiet window, say so in the scope and rules and the assessment follows it. You can also exclude any paths you do not want touched.

Link to section: Can I test a domain I do not own?Can I test a domain I do not own?

No. Assessments only run against domains you verify, and ownership is re-checked immediately before every run. You must own, or be explicitly authorized to test, every target.

Link to section: How long does an assessment take?How long does an assessment take?

Assessments typically take several minutes. You can leave the page and return - progress is saved and the report appears when the run completes.

Link to section: What happens if my domain verification lapses?What happens if my domain verification lapses?

If ownership can no longer be confirmed when you start a run, the target is marked for re-verification and the assessment does not start. Re-verify the domain to run again.