Skip to content
  1. Home
  2. AI
  3. AI permissions

What an AI app can see and do in your firm, and how to stop it

In Firmary, an AI app can do only what the person who connected it is allowed to do. An owner decides which AI apps may connect and how far each may go. For each kind of work the firm picks On its own, Asks first or Never. Sensitive actions can need approval in Firmary or a fresh sign-in, and every action an agent takes is logged where owners can see it.

By the Firmary team. Facts checked 12 November 2026.

The nine controls, and who sets each one

ControlWhat it stopsWho sets it, and where
The person’s role and client assignmentsAn AI app seeing clients or records its person can’tOwner or administrator, under People & permissions
The firm connection policyAny AI app connecting at all, when it’s turned offThe firm owner, under Agents › Access and policy
Limits for each AI app and profileOne app or profile doing more than you allowThe firm owner, under Agents › Access and policy
The rule for each kind of work: On its own, Asks first or NeverActions you want a person to confirm, or never want an agent to takeThe firm
Confirmation before any changeA change running before the person has seen it, unless they’ve set that tool to always allowThe person, in their AI app
Approval in Firmary, or a fresh sign-inSensitive actions, such as voiding an invoice or changing someone’s role, going through on a chat message aloneFirmary, for those actions
Never through AIClient EINs, API keys and connection management going through an AI appBuilt in. Nobody can switch it on
The audit logAn action happening with no record of itAlways on. Owners can see it
Suspend or revokeA connection staying open after you no longer want itThe person, under Agents › AI apps › Your connections, or an owner or a role you allow, under Everyone’s

What a connected AI app can reach

Client EINs never go to an AI app. Ask for one and it gets a link to the client in Firmary instead.

The screens you’ll use to control it

Product preview. Fictional client data.

The access screen when someone connects an AI app
An approval waiting in Needs review
The activity log
The firm’s switch for AI apps

What happens to the conversation

Firmary sends the AI app only the results of the requests it makes, and doesn’t receive the conversation. Your AI provider’s terms for your plan cover the conversation, so check them, including whether the provider can use it for training.

Ten questions to ask any vendor about AI and client data

Take these into any demo, ours included.

AskFirmary’s answer
Can the AI do more than the person using it?No. An agent can do only what its person is allowed to do.
Is every AI action logged, and can owners see the log?Yes to both.
Can AI be switched off for the whole firm?Yes. Your firm’s owner turns off AI apps under the firm connection policy, and every connection stops, including agents.
Can I require a person to confirm before an AI changes anything?Yes. Set that kind of work to Asks first, or to Never.
Is anything never available to AI?Yes. Client EINs, API keys and connection management.
Do sensitive actions need more than a chat message?Some need approval on a Firmary page or a fresh sign-in.
Does the vendor receive our AI conversations?No. Firmary gets only the requests the AI makes.
Do the vendor’s own AI providers train on our content?Not for Firmary’s built-in AI. Its providers work under terms that don’t allow it.
Can our clients connect their own AI to our data?A client can connect their own AI app to PocketFirm. It reaches only that client’s own account and sees only what they can see, and by default it asks before it changes anything.
Has the vendor completed a SOC 2 examination?No. Readiness work is under way, with a Type I report planned first, then Type II. Security and data handling lists the controls Firmary has today.

The rules your firm works under, whatever software it uses

Firmary doesn’t make you compliant with any of these. They’re the rules you check any tool against.

RuleWhat it coversRead it
IRC §7216 and Rev. Proc. 2013-14How a tax return preparer may use and disclose tax return information, and what a taxpayer’s consent to disclosure must sayUnder §7216, a preparer needs the taxpayer’s written consent to disclose or use return information, unless the regulations allow that use without it
IRS Publication 4557, Safeguarding Taxpayer DataThe IRS’s guide for tax professionals on protecting client data, including a written security planIRS Publication 4557
The FTC Safeguards Rule (16 CFR Part 314)Requires financial institutions, which include tax preparers, to run an information security program16 CFR Part 314

How to turn it off

Your firm’s owner can turn off AI apps for the whole firm. Each person can disconnect their own AI app, and an owner can revoke anyone’s. You can also run the whole firm in Firmary without connecting any AI app.

Questions about AI and client data

Is it safe to use AI with client data?

It depends on the controls a vendor gives you. Check whose access the AI uses, which changes need approval, what is logged and how an administrator can stop it.

Can an agent act without asking?

Only for the kinds of work your firm sets to On its own.

Can we turn AI off for the whole firm?

Yes. Your firm’s owner turns AI apps off under the firm connection policy.

Do AI model providers train on our data?

For Firmary’s built-in AI, the providers work under terms that don’t let them train on that content. For your staff’s own ChatGPT, Claude or Codex accounts, that provider’s terms for your plan apply.

Does using ChatGPT on tax returns need client consent under §7216?

It depends on how you use it. Under the §7216 regulations, a preparer needs the taxpayer’s written consent to disclose or use return information, unless the regulations allow that use without it. Ask your adviser how that applies to the AI apps your firm uses. This page isn’t legal advice.

Leave your work email.