Shadow AI: what it is and how to manage it

Learn what shadow AI is, why staff using unapproved AI tools puts customer data at risk, and how a small store can manage it without banning AI.

Shadow AI is the use of AI tools for work without the business’s approval or oversight. A store can manage it by finding the tools staff already use, approving suitable alternatives and setting clear rules for customer information. A paid account alone doesn’t make a tool safe for every task.

What shadow AI means

Shadow AI happens when someone introduces an AI tool into a work task without the store reviewing how it handles information. The tool might be a separate app, a browser extension or an AI feature inside software the team already uses.

For example, a staff member might paste a customer’s complaint into a personal ChatGPT account to draft a reply. Someone else might invite an AI note taker to a supplier call without approval. Both are shadow AI, even if the staff members only want to save time.

The same task can become approved use when the store reviews the tool, controls the account and specifies the information staff can enter. Approval needs to cover the task as well as the product. A tool approved for public product descriptions isn’t automatically approved for customer complaints.

Why shadow AI is a risk

Shadow AI can move information outside the store’s control before anyone responsible for it knows the tool exists. A customer email might contain an address, order number or details about a health condition. A supplier document might contain prices protected by a non-disclosure agreement (NDA).

The risks depend on what staff share and what the service does with it:

  • An overseas service might process customer information in another country or disclose it to other providers.
  • Uploading confidential supplier terms can breach a promise to keep those terms private.
  • AI-generated replies can invent product claims or describe a returns policy incorrectly.
  • Personal accounts can leave the store without a reliable record of the information shared or a way to manage access when someone leaves.

The Office of the Australian Information Commissioner (OAIC) recommends keeping personal information out of publicly available generative AI tools as best practice. Its commercial AI privacy guidance also addresses overseas processing and provider access to inputs.

Removing a customer’s name doesn’t always remove their identity. An order number or a detailed account of an incident can still identify them. For drafting practice, use a made-up complaint with invented details instead of copying a live customer record.

Finding the AI tools staff use

Find shadow AI by asking staff about their work, then checking the business systems you’re authorised to review. Include occasional use and free accounts. A list of paid subscriptions will miss both.

Send a short survey with these questions:

  • Which AI tools or AI features have you used for store work?
  • What task did you use each one for?
  • Is the account personal or controlled by the business?
  • What information did you enter, and did you connect the tool to files, email or calls?

Explain that you need an accurate inventory to choose useful tools. Give staff a private way to report a customer-data incident immediately, separate from the general survey.

On work devices, review browser extensions and their permissions. Review connected-app permissions in business accounts and business sign-in records where available. Look for note takers that join calls and tools connected to shared drives. Tell staff what you’re checking and keep the review within your workplace policies and authority.

For each tool, record its owner, account type, work task, connected systems and information entered. A hypothetical entry might read: “Personal writing assistant account; drafts replies; customer emails pasted manually; no store approval.” That entry needs an incident check as well as an approval decision.

Browser lists and login records won’t reveal every phone app or personal account. Ask staff to confirm the inventory and nominate one person to keep it current.

Shadow AI management without a ban

Manage shadow AI with a short approved-tool list and rules staff can follow during their normal work. Give the team business-controlled accounts where those tools need them. Review the actual plan’s privacy terms and settings before approving its use.

The OAIC’s workplace AI guidance recommends policies, staff education and active management of enterprise privacy settings. Buying business licences still requires decisions about data retention and provider access.

For each approved tool, write down the permitted task and the data it can receive. Name who checks the output and who handles a suspected incident. Set a review point when its features, permissions or terms change.

A small store’s starting rule could read:

Use the approved business account to draft from public product facts. Keep customer records and confidential supplier documents out of prompts and uploads. A staff member must check product claims and store policies before anything is published or sent. Report accidental sharing to the store manager immediately.

Use the AI policy generator to turn those decisions into a staff policy. Demonstrate one allowed task and one prohibited upload, then ask staff to show the difference. Keep a request process for useful new tools so the approved list can meet real work needs.

Customer data in unapproved tools

If customer data has entered an unapproved AI tool, stop further sharing and investigate the specific exposure immediately. Deleting the visible chat is one containment step. It doesn’t establish that every retained copy has gone.

  1. Record which account and tool received the data. Note what was shared, when it happened and whether a public link or connected app widened access. Keep the incident record restricted.
  2. Revoke unnecessary connections and disable sharing links. Request deletion through the provider’s controls and support process, retaining evidence of the request and response.
  3. Establish what the provider retained and who could access it. Assess the people affected and the possible harm, including whether the data contained credentials or sensitive information.
  4. Determine whether the Notifiable Data Breaches (NDB) scheme applies and assess notification duties. Escalate uncertain or serious incidents promptly to your privacy adviser or lawyer.

The NDB scheme applies to covered entities. Most businesses with annual turnover of $3 million or less are exempt from the Privacy Act, but exceptions apply. Use the OAIC’s small-business coverage guidance to establish your store’s position.

For a covered entity, suspected eligible breaches require a prompt assessment. Take all reasonable steps to complete it within 30 calendar days of becoming aware of the grounds for suspicion. If serious harm is likely and remedial action hasn’t prevented that risk, notification is generally required. The OAIC’s data breach response guide explains the assessment and notification process.

Don’t wait for that assessment period to end before containing the exposure. Record the response and the decision about notification, then fix the task that led to the upload. Give the staff member an approved way to draft the next reply without sending customer data to an unapproved tool.

Give your store its own shopping agent

Customers ask in their own words. Your agent finds, compares and checks out from your catalogue, right on your site.

See it in actionFree trial · Simple setup