TL;DR
Data security with AI is a real concern — and a solvable problem. This guide cuts through the fear, uncertainty, and doubt to give you the practical rules every Montana business owner should know in 2026.
**Don't put data into an AI tool that you wouldn't email to a third-party vendor.** That's the bar.
If you wouldn't put a customer's social security number in an email to an outside accountant, don't put it in ChatGPT. If you wouldn't share a confidential contract by email with a stranger, don't paste it into AI without the right protections.
The good news: with business-tier subscriptions and proper configuration, you can put a lot more into AI than you might think — including confidential customer data, financial records, and proprietary information. The point isn't to use AI less; it's to use it correctly.
**Free ChatGPT, free Claude, free Gemini:** Your inputs may be used to train future versions of the model. Your data could theoretically influence future model outputs (in aggregate, not directly attributable).
**Business/team tiers ($25-$30/user/month):** Your inputs are NOT used for training. Your data is encrypted in transit and at rest. You get admin controls and audit logs. Most have SOC 2 certification.
**Enterprise tiers:** All of the above plus signed BAAs (for healthcare), additional compliance certifications, dedicated support, and often the option of dedicated infrastructure.
For any business use, use at least the business tier. The $25-$30/user/month cost is trivial; the data protection difference is large.
**Healthcare (HIPAA):** Requires Business Associate Agreements (BAAs) with every AI vendor that touches PHI. OpenAI Enterprise, Anthropic Enterprise, and Microsoft Azure OpenAI all offer BAAs. Public ChatGPT does not.
**Legal (attorney-client privilege):** Use AI configurations with zero data retention and clear contractual protection. ABA Formal Opinion 512 (2024) provides current guidance. Montana Rules of Professional Conduct 1.1, 1.6, and 5.3 apply.
**Financial services (GLBA, FINRA):** Specific configurations and audit requirements. Microsoft Azure OpenAI and AWS Bedrock are typical choices.
**Tribal enterprises:** Data sovereignty principles apply. Often requires on-premises deployment or tribally-controlled cloud. CARE Principles for Indigenous Data Governance.
**Education (FERPA):** Student data has specific protections. Microsoft 365 Education with Copilot or Google Workspace for Education with Gemini are typical compliant choices.
Every business deploying AI should have a written AI use policy. Ours typically covers:
**Approved tools.** Which specific AI tools (and which tiers) employees may use for business purposes.
**Approved use cases.** What kinds of work AI may and may not be used for.
**Prohibited data.** What categories of data are NOT to be put into AI tools (e.g., 'no patient identifiers in any tool without a BAA').
**Output verification.** When AI output must be reviewed by a human before action, and by whom.
**Disclosure to clients/customers.** When AI assistance must be disclosed (this matters for legal, healthcare, and some consulting work).
**Incident reporting.** What to do if data is accidentally exposed.
We include drafting a written AI policy with every engagement.
Most AI security incidents are not technical — they're someone using the wrong tool for the wrong job. The fix is training:
- One 90-minute training session for the whole team when AI is first deployed
- A one-page written reference for what they can and can't do
- A clear path for questions ('when in doubt, ask before pasting')
- Periodic refreshers as new tools are added or policies change
We include this training in every deployment.
Into free ChatGPT, no. Into ChatGPT Team or Enterprise with proper configuration, yes — millions of businesses do this every day with appropriate guardrails.
That's why training and a written policy matter. Most accidental misuse can be remediated. The bigger risk is a culture that hides mistakes; the fix is making it safe to raise concerns.
Major AI vendors (OpenAI, Anthropic, Google, Microsoft) have security teams and incident response that exceed almost any small business's capability. They're not hack-proof, but they're hardened. Your bigger risks are usually inside your own organization.
Last updated March 15, 2026 · Written by Aaron Whitfield, Montana AI Consulting.
Ready to act on this?