30 Days Gen AI Risk Trial -Start Now
Skip to main content
Policy in practice · Practical playbook

Triage a sensitive-data submission to AI

First distinguish an attempted AI submission from data that reached the service. Preserve the evidence needed to make that distinction without asking the employee to repeat the upload or spread sensitive information into another system.

For Security operations, IT support and incident coordinators

Synthetic example

An employee reports an unexpected upload

A team member believes a spreadsheet with internal records reached an assistant. They saw a security message but are unsure whether it appeared before or after the file was submitted.

What you are working with

  • The approximate time, assistant, account and device used.
  • The original file location and a description of its sensitive categories.
  • The message shown and any existing provider or control evidence.

A safer approach

  • Stop further submissions through the uncertain route.
  • Preserve existing evidence in the approved incident record.
  • Have the incident owner assess transmission and the appropriate response.

Expected outcome: The case is recorded with a clear known-versus-unknown status and reaches the people authorized to assess exposure and response obligations.

Put it into practice

Work through the procedure

  1. Stabilize the workflow without repeating it

    Ask the user to stop related uploads and explain what happened in their own words. Record the time and submission path. Do not reproduce the event with the real document. If a demonstration is necessary later, use synthetic content under the incident owner's direction.

  2. Establish what the evidence actually shows

    Review available endpoint, application and provider records with authorized administrators. Distinguish a warning, a completed block and a confirmed upload. If evidence is incomplete, mark transmission as unknown. A screenshot of a policy message alone may not settle what the service received.

  3. Assess the information through its owner

    Ask the data owner to classify the content and identify affected business records. Record a minimized description rather than copying the entire file into a ticket. Escalate possible credentials or other urgent exposure through the organization's existing security response procedure.

  4. Hand off containment and record the outcome

    Give the incident coordinator the timeline, evidence sources, data description and unresolved questions. Provider deletion or access changes require the relevant administrator's review. Follow the established incident process for assessments and notifications; do not promise that deleting a conversation reverses transmission.

Evidence before approval

What to check before proceeding

1. Transmission status is evidence-based

Ready when
The record distinguishes prevented, transmitted and unknown outcomes with supporting sources.
If the check fails
Keep the status uncertain and assign the missing verification.

2. Evidence is retained with limited access

Ready when
Only necessary material is stored in the approved incident location.
If the check fails
Restrict the record and replace unnecessary copies with references where appropriate.

3. An incident owner accepts the handoff

Ready when
A responsible person owns containment, assessment and follow-up.
If the check fails
Escalate through the established response channel rather than closing the support ticket.

Common mistakes to avoid

  • Describing every DLP event as a confirmed breach before establishing whether information left the endpoint.
  • Asking the employee to paste the sensitive prompt into an unrestricted chat so more people can investigate.
Workforce AI Security

Evaluate this workflow with Aona

Where Aona can help

For supported workflows, Aona's policy events and configured controls may help explain the attempted action and observed response. Review them alongside the actual deployment and other available evidence.

What to confirm

Aona does not determine legal notification obligations or erase information from a third-party provider. A control event alone should not be presented as proof of receipt, deletion or complete incident containment.

Turning this policy into an operational rollout?

Discuss the teams, devices and AI tools in scope, who will own the policy, and which deployment and evidence requirements need to be met before rollout.

FAQ

Questions about this workflow

Ask the incident owner to direct next steps. Preserve necessary evidence and check the provider's applicable controls. Deletion may be part of the response, but it should not be represented as proof that all copies or prior processing disappeared.
Technical evaluation

Turning this policy into an operational rollout?

Discuss the teams, devices and AI tools in scope, who will own the policy, and which deployment and evidence requirements need to be met before rollout.

Triage Sensitive Data Sent to AI | Aona AI