AI risk testing before launch
Know where your AI is most likely to fail before users do.
I test customer-facing and internal AI systems for the failures normal testing misses: wrong answers, leaked data, broken rules, and promises your business never meant to make.
Founder-led testing from someone with 20 years of software security experience, including AI used in high-stakes healthcare workflows.
Who this is for
This is for companies using AI with customers, employees, or private data.
If a bad answer could create legal, financial, security, medical, compliance, or reputation risk, your AI needs more than normal QA testing.
AI that answers questions, handles complaints, or responds to users directly.
Systems connected to orders, records, customer history, or private information.
Healthcare, legal, finance, insurance, compliance, or other sensitive workflows.
Tools that approve, deny, escalate, summarize, route, recommend, or trigger actions.
They may look harmless, but they often touch policies, documents, and sensitive business context.
RAG systems that pull from internal documents, uploads, tickets, records, or web pages.
Teams preparing to ship and wanting a clear second look before real users arrive.
Systems already in production that have never been tested against real misuse patterns.
60-second AI risk check
Does your AI need a closer look?
Check every statement that applies. Some factors carry more weight than others — legal and data exposure are scored higher than general risk indicators.
Check any statements that apply to your AI system.
How this can work
Start small. Go deeper if the risk is real.
You do not need to start with a large engagement. The path begins with a simple risk check, then moves toward deeper testing only when it makes sense.
Risk check
Use the checklist above to see whether your AI has obvious exposure.
Free written assessment
Send a short description. I'll reply with the biggest risks I would look at first.
Red team assessment
I test your AI using real attack patterns and give you proof, findings, and fixes.
Starts at $5,000 / projectOngoing testing
As your AI changes, I keep testing so new features do not create new blind spots.
Starts at $4,000 / monthWhat paid testing includes
I try to break your AI, then show you how to fix what I found.
Scope the risk
I learn what your AI does, who uses it, what data it touches, and what a bad outcome would look like.
Run structured attacks
I test for prompt injection, jailbreaks, data leakage, hallucinations, over-permission, and broken guardrails.
Report clear fixes
You get a plain-language report with proof, risk level, business impact, and practical next steps.
These failures are already happening
AI systems can say, reveal, or approve things the business never intended.
The details change from case to case. The pattern is the same: the AI behaved in a way the company did not expect, approve, or control.
McDonald's chatbot answered coding questions
A customer support bot went viral for leaving its intended role and answering unrelated technical questions.
Fast Company Unauthorized promiseUnited's chatbot promised a refund
A customer pushed the chatbot into making a commitment the company had not clearly authorized it to make.
View from the Wing HallucinationLawyers cited cases that did not exist
AI-generated legal research created fake citations, leading to sanctions and public embarrassment.
ACC
Sean Yunt
Founder & PrincipalFounder-led AI risk testing
Hi, I'm Sean.
I spent 20 years breaking software for a living. Most recently, I led security testing for AI used by one of the largest health systems in the country, handling real patients, prescriptions, and medical decisions.
That experience taught me something simple: when AI fails in a serious environment, the risk is not theoretical. Real people, real data, and real business decisions are involved.
I started Black Diamond Consulting because most companies shipping AI have never had someone seriously try to break it before users get the chance.
AI risk guides
Plain-English guides and technical research.
Start with simple explanations of how AI systems fail. Go deeper when you want the technical details.
Read the AI risk guides
Plain-English articles on data leakage, prompt injection, hallucinations, guardrail failure, and more.
Technical researchReview the research library
Deeper reports on LLM security testing, model behavior, attack patterns, and production risk.
Not sure where to start?Request a free assessment
Send a short description of your AI system. I'll point out the risks I would check first.
Free · No call required
Want a second set of eyes on your AI risk?
Tell me what your AI does, who uses it, and what data it touches. I'll send you a direct written assessment of the risks I would look at first.