Key takeaways
- Nearly 40% of AI interactions involve sensitive data, so this situation is common, not rare.
- Speed matters more than certainty in the first hour — contain first, investigate the details after.
- A clear, blame-free reporting path is what actually gets incidents reported quickly.
- Not every incident needs a lawyer, but data involving health, financial, or identity information usually does.
- This is general information, not legal advice — a real incident needs a real attorney's read on your specific situation.
Someone on your team pasted a customer's information into a chatbot to speed up a task. It probably wasn't malicious, and it's more common than most owners assume — nearly 40% of AI interactions involve sensitive data, according to Cyberhaven's research on workplace AI use. What happens in the next hour matters more than what happened in the moment it was typed. This guide walks through exactly what to do.
The first hour: contain, don't investigate
The instinct is to figure out exactly what happened before doing anything. Resist that. The first priority is containing the exposure, not understanding every detail of it — you can reconstruct the full picture afterward.
| Step | What to do |
|---|---|
| 1. Stop, don't delete | Have the employee stop using the tool for now, but don't delete the conversation — you may need it to understand exactly what was entered. |
| 2. Get the basics in writing | What tool, what was entered, roughly when, and whether it went anywhere beyond the tool itself (e.g., was output shared further). |
| 3. Check the vendor's deletion options | Look up the tool's data retention and deletion policy, and submit a deletion request in writing even if you're unsure it will fully work. |
| 4. Notify the AI policy owner | Whoever owns your AI governance process (see our governance starter kit) needs to know immediately, not at the next scheduled check-in. |
| 5. Decide if legal counsel is needed | Use the table in the next section to judge the type of data involved. |
It's tempting to wait and see whether anything comes of it. Some situations carry notification deadlines that start running from the moment of exposure, not from when you decide to act — waiting costs you time you may not be able to get back.
When to call counsel
Not every incident needs a lawyer on the phone that day, but some categories of data make legal input close to mandatory. This is general guidance, not a legal determination for your specific situation — when in doubt, the cost of a short consultation is far lower than the cost of guessing wrong.
| What was entered | Typical next step |
|---|---|
| Internal notes, no identifiable customer or financial data | Log the incident, reinforce the policy — legal input usually not required. |
| Customer names or contact details | Log it, check the vendor's retention policy, consider a quick legal check depending on volume. |
| Financial account numbers, health information, or government ID numbers | Call counsel before deciding on next steps, including whether anyone needs to be notified. |
| Data covered by a specific industry regulation (healthcare, financial services, legal) | Call counsel immediately — industry rules often set specific timelines and requirements. |
Why a blame-free reporting process matters more than the policy itself
An incident response plan only works if people actually use it, and people only report mistakes when reporting doesn't feel like a confession. If the first reported incident at your company results in visible discipline, every future incident goes unreported instead — and an unreported incident is far more dangerous than a reported one, because nobody can contain what they don't know about.
Reserve any disciplinary response for a pattern of ignoring the policy after training, not for a single honest, promptly reported mistake. The goal of the policy is fast reporting, and fast reporting requires psychological safety.
Copy-ready incident log
Use this to capture the basics quickly, before memory of the details fades.
Who to tell, and in what order
An incident that stays known to one employee and nobody else doesn't get contained — it just stays hidden until it resurfaces somewhere worse. A short, clear notification order keeps the response moving without turning into a company-wide announcement over what might be a minor, quickly-resolved issue.
| Who | When |
|---|---|
| AI policy owner / manager | Immediately — this person coordinates everything else |
| Legal counsel | Same day, if the data involved falls into a higher-risk category (see table above) |
| Business owner / leadership | Same day for anything involving customer or financial data, even if legal isn't needed |
| Affected customer(s) | Only after legal input, if notification is warranted — don't decide this alone |
| Rest of the team | Not by default — only if there's a broader lesson worth sharing once the incident is resolved |
Notice that the customer sits near the end of this order, not the beginning. That's deliberate — reaching out to a customer before you understand what happened and what your legal obligations are can create more confusion than it resolves, and in some cases can complicate a notification requirement that has specific legal language attached to it.
Preventing the next one
An incident response plan handles the aftermath. Preventing the next incident depends on two things this guide doesn't replace: a clear written policy on what can never go into an AI tool, and an approved-tools list that gives employees a safe option so they're not guessing. Nearly 40% of AI interactions involving sensitive data is a symptom of employees not having clear rules — not a sign that your team is careless.
Use the incident, without naming the employee, as a short training example. A real, specific scenario makes the policy's data rules concrete in a way a bullet list alone rarely does.
Turning one incident into a smaller next one
Every incident, once resolved, is a chance to make the next one less likely and less severe. After closing out the incident log, take fifteen minutes to ask three questions: Was the tool involved on the approved list? Did the employee know the data rule that was violated? Was there a faster way this could have been caught? The answers usually point to one concrete fix — adding the tool to a restricted category, adding a specific example to training, or tightening a rule that was too vague to follow consistently.
One incident doesn't mean your whole AI policy failed. A single specific fix, applied consistently, does more good than a sweeping new restriction that makes every AI tool harder to use for everyone.
What to have ready before an incident happens
The businesses that handle these incidents well aren't the ones with no mistakes — they're the ones who decided the process in advance, when nobody was under pressure. Have these three things settled before you need them, not while you're in the middle of one:
- A named policy owner everyone knows to contact, not "someone in management."
- An attorney's contact information on hand, ideally someone already familiar with your business rather than a name you'd have to find under pressure.
- A copy of the incident log template above, saved somewhere your team can actually find it quickly.
None of this requires a formal incident response plan document, though larger or more regulated businesses may want one. For most small businesses, having these three things ready is the difference between a calm, fast response and a scramble that wastes the first, most important hour.
A brief mention of this process during onboarding or the next handbook rollout also helps — employees who know an incident process exists, and that it treats honest mistakes fairly, are far more likely to actually use it when they need to.
Keep the saved template and contact list somewhere accessible without needing to search for them — a pinned note in your team's shared workspace, or a printed copy in a manager's desk, works better than a file buried in a folder nobody remembers the name of.
Common mistakes
- Waiting to see if it matters. Some notification obligations run on a clock that starts at exposure, not at your decision to investigate.
- Disciplining the person who reported it. This guarantees the next incident goes unreported.
- Skipping the vendor deletion request. Even an imperfect deletion request creates a useful record and sometimes actually works.
- Handling anything beyond routine internal data without legal input. Health, financial, and identity data carry rules that vary by state and industry — guessing is expensive.
- Not having a policy before the first incident happens. See our AI acceptable use policy guide to put the rules in place before you need them.
◆ Small Business AI Kickstart
Get AI ready today.
Before it's too late.
yforest AI Labs comes to your company, trains your team, and ships your first tools.
FAQ
An employee pasted a customer's info into ChatGPT. Is this automatically a legal problem?
Not automatically, but it depends on what was entered and what laws or contracts apply to that data. That's exactly the kind of question to bring to an attorney rather than guess at — the answer changes based on the type of data and your industry.
Should the employee be disciplined for reporting a mistake?
Punishing an honest, promptly reported mistake teaches your team to stop reporting them. Save any disciplinary response for a pattern of disregarding the policy, not a single reported slip.
Can we delete data from an AI tool after the fact?
Sometimes, depending on the vendor. Check the tool's data retention and deletion policy immediately, and request deletion in writing even if you're not sure it will fully remove the data — the request itself becomes part of your incident record.
Do we need to tell the customer whose data was involved?
That depends on what data was involved and what laws apply, which is a question for an attorney. Don't decide this on your own or wait too long to ask — some situations carry notification deadlines.
How do we prevent this from happening again?
Pair the incident response with a written AI acceptable use policy and an approved-tools list, so employees have a clear, safe alternative instead of guessing which tools are fine to use.
Sources
This guide is general information, not legal advice. Have a qualified attorney review any policy before you adopt it.