Arizona rules small business owners should know before using AI with customer data
What not to paste into a chatbot, what Arizona's breach notification law requires, and the questions to ask any AI vendor before your team touches client information.
- Arizona law defines personal information as a name plus a specified data element such as a Social Security number, driver's license number, taxpayer ID, medical information, or biometric data.
- If a breach is confirmed, an owner must notify affected individuals within 45 days of that determination.
- A breach affecting a large number of individuals also requires written notice to the Arizona Attorney General and the Arizona Department of Homeland Security.
- Entities covered by HIPAA are exempt from the state notification law, but federal duties still apply to them.
It is a familiar pause. A staff member is drafting a reply to a client, the wording is not quite right, and the fastest fix is to paste the whole email into a chatbot and ask for a better version. The cursor sits there. The client's name, phone number, and the reason they called are all in that text. Nobody in the office is sure whether pasting it is allowed. This piece gives the ground rules for that decision, drawn from Arizona's breach notification law and the confidentiality duties a business already carries.
What Arizona counts as personal information
Arizona's data breach notification law, A.R.S. §§ 18-551 and 18-552, spells out what counts as personal information. The law defines it as an individual's first name or first initial and last name, combined with at least one specified data element. Those elements include a Social Security number, a driver's license number, a taxpayer ID, medical or mental health information, and biometric data. The definition also covers a username or email address combined with a password or security question that opens an online account. Information that is publicly available from government records or widely distributed media does not count.
That definition is the practical line for an owner deciding what to paste into a chatbot. A prompt that says "write a follow-up email to a homeowner about a roof quote" is a task description. A prompt that includes the homeowner's name with a driver's license number or medical detail is personal information in the legal sense. The **Arizona Attorney General's Office** publishes a plain-language FAQ on the law, and it is worth reading before writing a team policy.
What the notification law requires
The law applies to any person or business that conducts business in Arizona and owns, maintains, or licenses unencrypted and unredacted computerized personal information. If that business becomes aware of a security incident, it must investigate promptly to determine whether a breach occurred. If the investigation confirms a breach, the owner must notify the affected individuals within 45 days of that determination, subject to law enforcement needs.
Two larger duties ride along. If the breach requires notifying a large number of individuals, the owner must also notify the three largest nationwide consumer reporting agencies, and the Attorney General and the director of the Arizona Department of Homeland Security in writing. One more point matters for clinics: entities covered by the federal Health Insurance Portability and Accountability Act, known as HIPAA, are exempt from the state notification law. That exemption does not remove the federal duties. It means a covered clinic answers to HIPAA rules instead of the state statute.
What not to paste into a chatbot
The safest habit for a small business is a simple one: describe the task, do not paste the person. A prompt can carry the shape of the work without the identifying details. "Draft a reminder to a client that their appointment is Thursday" gives a model everything it needs. The client's name, address, insurance number, or diagnosis does not need to travel with it. A real estate agent can ask for a counteroffer email without the buyer's financial details. A professional services firm can ask for a cleaned-up engagement letter without the client's tax ID.
There are two reasons this habit holds up. The first is legal. Arizona's statute turns on unencrypted personal information sitting in a computerized system, and a free chatbot account is a third party's computerized system. The second is professional. Clinics carry medical confidentiality, real estate agents carry duties to clients, and professional services firms carry their own client confidentiality obligations. Those duties do not pause because a tool is convenient. An owner who never puts client data into a tool that was never vetted for it has closed the easiest route to a breach.
Questions to ask any AI vendor
Before a tool touches anything sensitive, a careful owner asks the vendor a short set of questions and writes down the answers. The specifics will vary by tool, but the questions are the same across the market.
- 01Where does the data go?
Ask whether prompts and files sent to the tool are stored by the vendor, for how long, and on whose servers.
- 02Who trains on it?
Ask whether the vendor uses customer inputs to train its models, and how to turn that off.
- 03Is there a business associate agreement?
If the business is a covered clinic, ask whether the vendor will sign a HIPAA business associate agreement. No agreement, no protected health information in the tool.
- 04What happens in a breach?
Ask how the vendor notifies customers of a security incident, and what its contract says about liability.
- 05Can we delete it?
Ask how to delete stored prompts, files, and account data, and whether deletion is verified.
- 06Can we turn features off?
Ask whether chat history, training use, and integrations can be disabled at the account level.
The answers should land in a short written policy the whole team can follow. A one-page rule works better than a long one. It should say what may go into AI tools, what may not, which accounts are approved, and who to tell if something goes wrong. A staff member who knows the rule does not have to guess at the cursor.
How to set safe-use rules for a team today
An owner can build the policy in an afternoon without buying anything. The steps below work for a clinic, a real estate office, or a professional services firm, and they can be done without any outside help.
- 01List what counts as sensitive
Name the categories your business holds that match Arizona's definition, such as medical information, Social Security numbers, taxpayer IDs, and account credentials.
- 02Write the no-paste list
Put those categories in one short list and tell staff none of them go into an AI tool without an owner's sign-off.
- 03Teach the describe-the-task habit
Show the team how to write prompts with placeholders, like "the client" or "the property," instead of real details.
- 04Pick approved accounts
Decide which AI accounts the business uses, who holds them, and what settings they run with.
- 05Name a breach contact
State in the policy who staff tell first if data leaks, so the 45-day clock starts with someone in charge.
- 06Revisit it quarterly
Set a calendar reminder to reread the policy and the Attorney General FAQ, since tools and settings change.
That is the whole shape of the work. Arizona's law gives the deadlines and the definition of personal information. The owner's job is to keep the sensitive material out of unvetted systems in the first place, and to be ready to investigate and notify fast if something slips anyway. A team with a written rule and an approved account setup is a long way ahead of one where every staff member improvises.
