Services / Packages
AI Red Team Assessment
Adversarial testing that tries to make your AI system leak data, break its rules, or produce harmful output.
From $7,500 per system2–3 weeks
Public agencies receive quotes structured to fit their procurement rules.
You own everything we produce.
Who this is for
Organizations usually call us about this when:
Your AI system is open to the public or connected to sensitive data.
Security leadership wants to know how the system holds up against misuse.
You need adversarial testing evidence for a security review or audit.
Overview
Red teaming tests how an AI system holds up against people trying to misuse it. We attempt prompt injection, jailbreaks, data extraction, and manipulation to produce harmful, false, or off-policy output, then document what worked, how serious it is, and how to fix it.
What’s included
- Threat model for the system and its users
- Prompt injection and jailbreak attempts
- Attempts to extract confidential or personal data
- Tests for harmful, biased, or off-policy output
- Findings rated by severity, with fixes
- Retest of fixed issues and a summary for leadership
Available under attorney-client privilege. We can perform this work at the direction of your counsel, so findings may be protected by privilege. Ask us about privileged engagements.
How it works
Model threats
We identify who might misuse the system, how, and what they could gain.
Attack
We run adversarial tests and document every successful technique.
Fix and retest
We recommend fixes, retest them, and summarize the results.
How it works for your organization
The same service, tailored to the obligations and realities of each sector.
Private Sector
Companies use red teaming to find weaknesses in AI products and assistants before attackers or customers do.
- Public-facing AI assistants and agents
- AI connected to internal data or tools
- Protection of customer and business data
- Evidence for security reviews and audits
Public Sector
Agencies use red teaming to make sure public AI tools can’t be manipulated into exposing data or misleading residents.
- Resident-facing chatbots and portals
- AI with access to case or records systems
- Misinformation and impersonation risks
- Findings that support security reviews
Nonprofits
Nonprofits use red teaming to protect client and donor information in the AI tools they adopt.
- Chatbots that handle client questions
- AI connected to client or donor databases
- Harmful or inappropriate output risks
- Practical fixes for small teams
Frequently asked questions
How much does AI Red Team Assessment cost?
AI Red Team Assessment starts at $7,500 per system. The final price depends on scope, and we confirm it in writing before any work begins.
How long does it take?
Most engagements take 2–3 weeks, depending on scope and how quickly information is available.
What do we receive?
The engagement includes: threat model for the system and its users; prompt injection and jailbreak attempts; attempts to extract confidential or personal data; tests for harmful, biased, or off-policy output.
Can this be done under attorney-client privilege?
Yes. When your counsel engages us and directs the work for the purpose of giving legal advice, our findings may be protected by privilege. Ask your counsel to contact us and we’ll structure the engagement with them.
What if a deliverable doesn’t meet the agreed scope?
We revise it until it does, at no extra cost. And if we miss an agreed delivery date, we take 10% off that deliverable.
Who owns the deliverables?
You do. Everything we produce for you is yours to keep, use, and share.
Related Guides
- AI Impact Assessment Example for Credit UnionsA worked example assessing an automated personal loan decision model, from purpose and data to risks, controls, and approval.
- AI Risk Register for Credit UnionsA sample register of common credit union AI uses, the data each one touches, its risk tier, the controls it needs, and who owns it.
- AI Tabletop Exercise Scenario for Credit UnionsA ready-to-run scenario where a member service chatbot exposes account details, with timed developments and questions for each phase of the response.
- AI Laws by StateWhere each state’s AI and privacy rules stand, with key dates and a page for every state.
Related services
AI Assurance Test
Our most thorough test: an AI system run against 100 or more realistic scenarios, with a signed assurance letter.
AI Agent Go-Live Review
A fast pre-launch check of an AI agent’s permissions, guardrails, and escalation, lighter than a full assurance test.
AI Risk Review
An independent review of one AI tool, vendor, or use case before you adopt it or expand it.
Talk through your AI and privacy questions
Tell us what you're working on, and we'll reply within one business day.