Will it leak your data?
Client records, contracts, and notes pasted into a public chatbot. An app connected to email or files. A website bot talked into giving up something it can see. The question is whether private information stays private.
How we work · Houston and remote
QUESTIONS / STEPS / SCOPE
Three questions
about your AI.
A checkup asks whether AI will leak your data, whether it can be tricked into a promise you have to keep, and whether it will stay up. Then it follows the same six steps.
Written for an owner. The framework names are cited below. This page is not a certification and not legal advice.
Can it be tricked?
A bot that invents a refund, a price, or a policy, which you may then have to honor. A cloned voice that gets a payment approved. The question is whether you can stand behind what the AI said or did.
Will it stay up?
A chatbot that is down when a customer needs it. One that someone keeps prompting until the bill spikes or the tool is useless. The question is whether it stays available and resists that abuse.
The labels are confidentiality, integrity, and availability: the CIA triad, a standard security checklist. CIA here means that checklist, not the agency. The words are the labels. The questions are the checkup.
01 / The process
Six steps, same order.
Mapped to the six functions in NIST CSF 2.0. The function names come from that framework. A checkup follows this order.
NIST CSF 2.0 is a public cybersecurity framework from the National Institute of Standards and Technology. Version 2.0 names six functions: Govern, Identify, Protect, Detect, Respond, and Recover.
What you can start today is the AI risk checkup, and scam and account protection. A written staff AI policy and a red-team test of a chatbot belong in this method. They are not a separate offer yet. The services page lists them as coming next while the training continues.
Govern. A policy for staff.
An AI use policy: which tools people may use, what they may not paste in, and who can approve a new app. The checkup asks how staff use AI today. The written policy is the Govern step.
Identify. The tools and the data.
An inventory of the AI already in the business. Website chatbots, office copilots, apps connected to email and files, and the customer or company data each one can see. What is not on the list is hard to protect.
Protect. Limits that hold.
Controls that limit the damage. Access limits, so a tool only reaches what it needs. Filters on what goes in and what comes back out. Callback rules for payment requests, so a familiar voice is checked before money moves. Email spoofing protection and two-step login sit here too.
Detect. Testing, with permission.
Checks that the protections are actually in place, including the free domain checkup, which reads only public information. For a chatbot the business runs, the method includes an authorized red-team test: a planned attempt to trick the bot, using the OWASP Top 10 for LLM Applications (opens in a new tab) as the checklist. That test happens only inside a written scope. The tools for it are still being built.
Respond. Findings, in order.
Clear findings with fixes, ranked by priority. One report. Each area marked red, yellow, or green, with the fix in plain language and the order to do them. The same grades as the free checkup. The report says what was checked and what was found. It is not a guarantee and not a certification.
Recover. A plan and a recheck.
A plan if something has already gone wrong, plus a recheck. Who to call, what to turn off, and what to tell customers if a payment went out, an account was taken, or a bot published a bad answer. Then a second look to confirm the fix held.
02 / Standards
Built on public frameworks.
Citations, not a certification. NIST, OWASP, and MITRE have not reviewed this work, and there is no affiliation with them.
These four documents are the references. They are free to read. Naming them means the method can be compared with something public.
NIST CSF 2.0The Cybersecurity Framework, version 2.0. The six functions above come from here: Govern, Identify, Protect, Detect, Respond, Recover. (opens in a new tab)
NIST AI Risk Management FrameworkA separate NIST guide for risk in AI systems. Used as a reference for how AI risk is described, not as a certificate. (opens in a new tab)
OWASP Top 10 for LLM ApplicationsA public list of the common ways language-model apps fail, including leaks and being talked into a bad answer. Used as the checklist when a chatbot is tested. (opens in a new tab)
MITRE ATLASA public catalog of real cases in which AI systems were tricked or abused. Used as a reference for those cases, not as a product. (opens in a new tab)
No certification is claimed. No official affiliation with NIST, OWASP, or MITRE is claimed.
03 / Always authorized
Testing only with written permission.
A checkup, a closer look at a system you run, or a test that tries to trick a chatbot: none of it starts without your written permission and an agreed scope.
The scope names what will be checked, what is off limits, and when the work stops. A red-team test is included only if the scope says so. If the scope is unclear, the test waits.
The free domain checkup only reads public information. The form asks you to scan a domain you own or have permission to check. Results are not stored.
Written scope. Then the work.
No surprise tests. Nothing outside the agreement.
04 / About
Houston-based.
Houston-based.
The lab is public.
There is no client list yet. This page does not fill the gap with testimonials or statistics.
Raúl Wesche is based in Houston. He runs his own local AI hardware, quantizes open models, and red-teams them: planned tests for ways they fail. The experiments are published at wesche.com/lab, including the ones that fail.
He is working toward the Google Cybersecurity Certificate, and then CompTIA Security+. Neither is finished. The background section keeps the same record, including training that is still ahead.
05 / Contact
Xx.com/WescheNex1q (opens in a new tab)
LinkedInlinkedin.com/in/raul-wesche-ab1320443 (opens in a new tab)
Tell me about the business.
Send a message on X or LinkedIn: what you do, which AI tools you or your staff use, and what worries you. That is enough to say whether a checkup is the right first step.
Security testing by Raúl Wesche, Houston. This page is not a law firm and not legal advice.