Compliance

AI usage policy for law firms and dental offices

How law firms and dental offices can stop staff pasting client data into ChatGPT, with policy and managed browser controls from Obsidian Ridge.

SMB

Short answer: Give staff a written list of approved AI tools, accounts and information they may use. Apply browser controls to enforce the supported rules, then test them with invented examples. A policy or browser block alone does not cover every desktop agent, personal account or unmanaged device. Obsidian Ridge's Free 30-day AI security assessment shows which AI tools and agents your staff already use before you write the list, with no charge and no card.

Key facts

  • Can a policy alone stop a paste? No, staff guidance needs technical controls and checks of the actual work paths.
  • Approved use: Record the tool, account, permitted information, task and responsible owner.
  • Client data: Removing a name alone does not establish anonymity or permission to share.
  • Browser scope: Island can restrict supported data movement; AI Protect is a separately scoped add-on.
  • Mistakes: Give staff a known reporting route and preserve the relevant facts.

Free 30-day AI security assessment

Obsidian Ridge runs a free 30-day AI security assessment for small businesses. It finds AI apps and agents on staff devices, what they can reach and risky actions. A CISSP reviews the findings and recommended fixes with you. There is no charge and no card. It is built for any size, from a one-person practice up, and implementation and paid tools are separate.

Request my free assessment

How can a small law firm or dental office stop staff from pasting client data into ChatGPT and other AI tools?

At Obsidian Ridge, we combine an approved-tool policy and staff training with managed Island Enterprise Browser controls that can block client-data paste into public AI tools in the supported, configured work environment. Test the allowed and blocked paths with dummy data and confirm any separately priced Island AI Protect features before rollout; personal apps and devices outside that scope need their own controls.

What should the policy protect?

For lawyers, ABA Formal Opinion 512 discusses professional duties when using generative AI, including competence, confidentiality and communication. Approving a tool does not approve every client matter or every data use.

For HIPAA-regulated practices, a service that creates, receives, maintains or transmits ePHI on the practice's behalf can involve business-associate obligations. HHS explains that an appropriate BAA and compliance with applicable safeguards are needed for covered cloud uses. A consumer account should not be treated as approved simply because it offers a privacy setting. HHS cloud guidance.

Copyable one-page policy

Approve this with the people responsible for privacy, professional obligations and IT. Maintain the approved-tool register and reporting contact separately so staff can find current details.

To fill in the gaps as you read it, the staff AI policy builder assembles these same clauses with your approved tools and information types, in your browser.

STAFF USE OF GENERATIVE AI

Purpose and scope
This policy applies to work performed for the practice using AI websites,
applications, browser extensions, coding assistants and automated agents.
It applies on work devices and whenever practice information is involved.

Approved tools and accounts
Use only tools and work accounts in the practice's approved-tool register.
Approval names the permitted tasks, data, settings and responsible owner.
Do not connect personal AI accounts, browser extensions or agents to work
email, files, clinical systems or case-management systems without approval.

Information restrictions
Do not enter client, patient, taxpayer, payment or authentication information
into an unapproved tool. Do not upload matter files, clinical records,
recordings or screenshots containing that information. Removing a name alone
does not establish that information is anonymous or properly de-identified.
Use public or invented examples for experimentation.

Permissions and actions
Request approval before granting an AI tool access to mail, drives, calendars,
local files or other systems. Use the least access needed for the approved
task. An agent must not send communications, change records or make purchases
without the authorized person's review and approval.

Review and professional responsibility
Check AI output for accuracy, sources and inappropriate disclosure before
using it. A qualified person remains responsible for professional work.
Follow any client-consent, patient-privacy and contractual requirements.

Mistakes and concerns
Report a wrong upload, unexpected access or unintended action promptly to
the practice's designated reporting contact. Preserve the relevant facts;
do not delete records or continue experimenting to investigate alone.

Ownership and maintenance
The practice's designated owner approves exceptions in writing, records
training and updates this policy when tools, tasks or obligations change.
Staff must know where the approved-tool register and reporting contact live.

Which controls Obsidian Ridge can use to enforce it

The managed browser supplies policies for work-app access and supported data movement. Island describes AI Protect as extending controls across its supported browser, extension and other components; that product description does not establish what is included in a base-browser deployment. Confirm the supported sites, allowed actions, work-account rules and exception process in the proposal. Island AI Protect overview.

That does not prove coverage of a desktop AI application, a command-line agent or a personal device outside management. The Free 30-Day AI Security Assessment examines the agreed device and AI-tool scope. Findings still need an owner, a decision and a record of the changes made. See the AI assessment comparison before choosing a browser control, discovery tool or assessment, and what each AI vendor publishes about its own tool before you name one in the policy.

How should staff learn the policy?

Use examples from the practice: a client letter, a radiograph screenshot, a bank detail and a public marketing paragraph. Ask which tool and data combination is approved and where an exception goes. Keep completion records and repeat the exercise when the tools or rules change.

Managed training can support the programme, but the practice must approve its policy and decisions. A technical block should explain the permitted next step, not leave staff to invent a workaround.

What should happen after an accidental upload?

Follow the incident process, preserve the relevant tool, account, time and information details, and involve the authorized privacy/security contact. That person can coordinate containment and assess contractual, professional and legal duties. Do not promise that deleting a conversation removes every copy or reverses access already granted.

For the broader programme, use the law-firm guide or dental programme. Our pricing separates ongoing controls from assessment and evidence work. Agree the systems and tools the policy needs to cover.

What this costs with us

Free 30-day AI security assessment

Obsidian Ridge runs a free 30-day AI security assessment for small businesses. It finds AI apps and agents on staff devices, what they can reach and risky actions. A CISSP reviews the findings and recommended fixes with you. There is no charge and no card. It is built for any size, from a one-person practice up, and implementation and paid tools are separate.

Request my free assessment

See current pricing for the full scope and separate charges.

Sources

Fetched October 2, 2026.

Last updated

October 31, 2026. We refresh this content as the threat landscape and tools evolve.

Questions readers usually ask next

Can a policy alone stop a paste?

No, staff guidance needs technical controls and checks of the actual work paths.

How can a small law firm or dental office stop staff from pasting client data into ChatGPT and other AI tools?

At Obsidian Ridge, we combine an approved-tool policy and staff training with managed Island Enterprise Browser controls that can block client-data paste into public AI tools in the supported, configured work environment.

Test the allowed and blocked paths with dummy data and confirm any separately priced Island AI Protect features before rollout; personal apps and devices outside that scope need their own controls.

Does removing a patient or client name make AI use safe?

No. Removing a name does not prove that the remaining information is anonymous or properly de-identified. Apply your approved data rules and seek appropriate privacy or legal review before sharing sensitive information.

Is Island AI Protect included in the browser price?

Island AI Protect is a separately priced add-on. The published standalone browser rate is for Island Enterprise Browser. Ask for the add-on quote and supported enforcement scope.

Is this template legal advice or proof of compliance?

No. It is an operational starting point that the practice must approve and adapt. It does not replace professional duties, a HIPAA analysis, client terms or legal review.

What should staff do after an accidental upload?

Follow the practice's incident process and promptly contact its designated privacy or security owner. Preserve the tool, account and information details without assuming that deleting the chat removes every copy.

Full bio & provenanceSee related service