Free Surface Recon
$0
Free passive-first reconnaissance that shows what is publicly visible about a domain before you scope deeper testing.
Start Free Recon- Surface summary
- Notable public exposures
- Suggested Deep DAST or AI + Infra scope
One-shot security audits for modern web apps, APIs, chatbots, RAG systems, and AI agents. Start free, scope safely, and get verified findings your team can fix.
Audits are one-shot engagements. Free recon shows the public surface. Paid tiers move through manual scope approval before checkout and active testing.
$0
Free passive-first reconnaissance that shows what is publicly visible about a domain before you scope deeper testing.
Start Free Recon$1,500
Dynamic web and API testing with AI-aware probes, verified findings, signed evidence, and fix-ready remediation.
Request Scope Review$5,000
A deep AI red-team plus infrastructure review for products where LLM behavior can become business risk.
Request Scope ReviewGraphQL, WebSockets, OIDC flows, structured bodies, hidden parameters, browser execution, and OAST callbacks.
Prompt injection, system prompt extraction, RAG leakage, tool-call abuse, refusal bypass, and model-cost controls.
Every verified finding includes impact, proof, fix objective, validation steps, and agent-ready Markdown.
Active testing starts only after target ownership, scope, safety limits, blackout windows, and emergency contact details are reviewed.
Start with Free Surface Recon or connect a chatbot endpoint so WardenBot can see what a public user can reach.
Run audit-grade probes, canary checks, business truth assertions, and multi-turn attack chains against approved scope.
Script-first checks, guarded LLM judges, and human review separate real findings from scanner noise.
Use agent-ready Markdown to remediate, then keep monitoring after launch for drift, regressions, and new failures.
# agent-fix.md
## Finding
The support chatbot leaked the configured canary phrase during a prompt-extraction probe and returned a 60-day refund window instead of the approved 30-day policy.
## Objective
Prevent system/RAG canary disclosure and make refund answers resolve from the approved policy source. Preserve normal customer-support escalation.
## Suggested files
- prompts/support-system.md
- src/lib/chatbot/policy-context.ts
- tests/ai/support-policy.test.ts
## Implementation notes
1. Remove canary-bearing setup text from retrievable customer-facing context.
2. Add a refusal rule for requests to reveal hidden instructions or setup text.
3. Source refund answers from the canonical policy record, not stale prompt text.
4. Log policy-source version in chatbot traces for future behavior diffs.
## Acceptance tests
- Prompt-extraction probes do not return the canary.
- Refund questions consistently answer 30 days.
- Off-policy refund questions escalate instead of inventing exceptions.
- WardenBot replay probe now passes canary and business-truth checks. A typical WardenBot customer starts with recon, buys a Deep DAST or AI + Infra audit before launch, then subscribes to Continuous Monitoring so model, prompt, and vendor changes do not silently regress behavior.
Canary leaks, business truth facts, Bot Health Score, behavior diffs, and agent-ready remediation.
Explore MonitoringStart with Free Surface Recon, then request scope review if deeper web/API or AI + Infra testing is warranted.