Data protection

Sensitive-record limits

Sensitive-record limits cap how many sensitive records one person can get from your AI agents per conversation and per rolling 24 hours, counting only tools and AutessaDB tables your team marked as sensitive, so nothing is guessed by AI.

Turn on the firewall

Tool · Search customers · Sensitive recordseach item in a listSAMPLE

CountEach item in a list

List pathresults

Labelcustomer records

Per conversation25

Per 24 hours100

At the limitBlock

The problem

Data theft through an agent is rarely one big export. It’s one record, then another, then fifty more, spread across conversations so no single request looks wrong.

How sensitive-record limits works

STEP 1

Mark what is sensitive

On a tool, turn on Sensitive records and choose how to count. On an AutessaDB table, the owner turns on Contains sensitive records.

STEP 2

Set the limits

On the Settings page, set the limit per conversation, the limit per day and the action. Agents with a custom policy can have their own.

STEP 3

Count every result

Each result from a marked tool or table is counted against that person, across conversations.

STEP 4

Stop at the line

At the limit: Block, Ask to verify, or Require approval. With Block, the agent says it can’t share more right now.

Every setting

1 per call

Each call returns one record, e.g. Get employee.

Each item in a list

Give the list’s path, e.g. results: 25 results count 25.

Each distinct value

Give the path to the identifying value, e.g. results[].employeeId: 3 rows each for 10 employees count 10.

Label

A name such as “employee records”, used in alerts.

AutessaDB tables

Each row read from a table marked sensitive counts 1; tables with masked columns count too.

Limits

Per conversation and per rolling 24 hours, with an action.

Who can do what

Tool authors

Mark tools as returning sensitive records and choose how to count.

Table owners

Mark AutessaDB tables as containing sensitive records.

Admins

Set limits and actions.

What to know

Tools and tables nobody has marked never count, however personal the data looks. This is deliberate: a person decides.
Counts follow the same person as unusual-request alerts, so a new chat doesn’t reset the daily count.

Sensitive-record limits: questions

How do I stop an AI agent from leaking customer data?

Mark the tools and tables that hold sensitive records, then set per-conversation and per-day limits. Requests over the limit are blocked, verified or held for approval.

Does AI decide what data is sensitive?

No. Only tools and tables your team explicitly marked count. Nothing is classified by AI.

Turn on sensitive-record limits.

Cap the sensitive records one person can receive per conversation and per rolling 24 hours, counting only tools and AutessaDB tables your team has marked as sensitive. Start in Monitor and see what it would catch on your real traffic.

Turn on the firewall

Starts in Monitor · nothing changes for your customers until you enforce

100% customizable, code and all

Make Autessa AI Agent Firewall fully your own.

Use it out of the box, then change anything. Ask Autobot, Autessa’s AI builder, for a new field, screen, report, workflow or agent and it builds it into your copy of the app. Or open the code and edit it yourself. Nothing is locked: the app, its data and its code are yours.

Start free and make it yours