Security comes first.
AI gives attackers new ways in. Here is every step we take, and the standard behind it.
Nine steps, in order.
Each one names its standard, so your IT partner can check it.
We map your data first.
We list every system the AI will touch and what is in it. Anything the job does not need stays out of reach.
NIST CSF 2.0: IdentifyThe AI gets its own login, with the least access it needs.
It never borrows a person's password. Each AI key has an owner and a spending cap. Admin accounts use security keys or passkeys.
NIST CSF 2.0: Protect · CISA phishing-resistant MFAA person approves anything that leaves the building.
The AI drafts quotes, certs, and replies. It cannot send them, or change a price, date, shipment, or payment, unless you approve that task in writing.
OWASP LLM Top 10 2026: excessive agency · CISA agentic AI guidanceEvery incoming document is untrusted.
A PDF or email can hide text written to trick an AI. The system treats documents as information only, with a short list of allowed actions.
OWASP LLM Top 10 2026: prompt injectionYour data is never used to train AI models.
We use business AI services that forbid it, named in our privacy policy. Data is encrypted in transit and at rest.
NIST AI 600-1: data privacyEvery action is logged.
What the AI read and drafted, who approved it, and when.
NIST CSF 2.0: DetectWe test it, then try to break it.
It must clear your bar on 50 past jobs. We also feed it planted instructions and wrong values to check that it flags them.
NIST AI RMF: MeasureWe vet every add-on.
A short, named supplier list, each one reviewed, with versions pinned so nothing updates itself.
NIST CSF 2.0: Govern · OWASP: supply chainThere is an off switch and a plan.
Each AI login turns off in one step. If something goes wrong, we stop, tell you in writing, fix the cause, and retest.
NIST CSF 2.0: Respond, Recover
Where we stand today.
We do not hold SOC 2 or ISO 27001 yet. We use their controls as our checklist and tell you which ones each system meets.
If CMMC applies to you, the AI keeps controlled information inside your approved environment, and we show your assessor where each part sits.
First Pass files are used only for that draft and deleted within 90 days. NDA on request.
Stop at any checkpoint. The AI's logins are turned off and your records stay yours.
The standards we build to.
Questions we hear
Are you SOC 2 or ISO 27001 certified?
Not yet. We use their controls as our checklist and tell you which ones each system meets.
Is our data used to train AI?
No. Our AI providers' terms forbid it; they are named in our privacy policy.
Can the AI send an email or change a price on its own?
Only for a low-risk task you approve in writing. Otherwise a named person approves it.