AI Agent Red-Team Assessment
Find your agent's weaknesses. Before someone else does.
A fixed-scope red-team assessment of your AI agent. We attack it the way an adversary would: prompt injection, prompt extraction, tool misuse. Then we hand you a report with scoped findings, available evidence and clear limitations. Confirm scope and scheduling before booking.
Written authorization · Agreed scope · Synthetic data first
Before you book
Assessments are custom engagements. Agree ownership, third-party permissions, test limits, deliverables, scheduling and retest terms with us in writing. Sales confirms the scope, price and applicable taxes before booking. No assessment certifies security or covers every possible attack.
What's included
OWASP LLM & Agent attack categories
Select relevant tests informed by OWASP risk categories. Agree the endpoints, available probes and exclusions; category mapping is not exhaustive attack coverage.
Prompt-extraction tests
Targeted attempts to extract your system prompt, hidden instructions, and internal configuration.
Tool-misuse detection with trace evidence
Review scoped tool behavior and available execution evidence. Trace detail depends on instrumentation and capture settings; missing evidence is identified.
Branded PDF report
A scoped report with severity-ranked findings, available reproduction evidence and explicit incomplete or untested outcomes.
Remediation guidance
Review fixes relevant to observed findings, such as prompt changes, application guards and tool permissions.
Agreed follow-up verification
Agree the retest scope, endpoints and timing with your assessment before follow-up verification.
How it works
Share a staging endpoint + written authorization
You give us a staging deployment of your agent and a signed authorization to test it. We only test what you own and authorize, in writing.
We run the assessment
Run the agreed probe set against the authorized endpoint, preserve available evidence and identify failed or incomplete checks.
Agree the delivery window
Confirm the reporting schedule before booking, then review scoped findings, reproduction evidence, limitations and remediation guidance.
Agree the scope. Request a quote.
A separate professional service under an agreed scope, price and schedule.
Launch
One agent, one report.
- 1 agent (staging endpoint)
- Agreed OWASP-informed test scope
- Branded PDF report
- Remediation guidance
- Agreed retest scope and timing
Standard
For teams shipping multiple agents.
- Up to 3 agents
- Agreed OWASP-informed test scope
- Branded PDF report per agent
- Remediation guidance
- Findings walkthrough call
- Agreed retest scope and timing
Agency Pack
White-label, for agencies.
- White-label report with your branding
- Resell to your clients
- Up to 3 agents
- Available trace and reproduction evidence
- Findings walkthrough call
- Agreed retest scope and timing
Questions? support@nyraxis.io
Questions, or need a custom scope? Email support@nyraxis.io. Assessments are performed only against systems you own or have written authorization to test.