Data Pasted Into an AI Tool: Your Response Checklist
An employee pasted client details into ChatGPT (or Claude, or another LLM) to finish a task faster, and you found out afterward. That sinking feeling is reasonable, and the situation is manageable with a clear order of operations. This checklist is for office managers and ops leads who need to act in the first hour and then decide what comes next. Visibility into which tools your team uses is part of what 24hourtek covers through AI Service & Shadow AI Monitoring for Business, and the sections below start with what happened to the data.
Key Takeaways
Depending on the provider and plan, data pasted into a public AI tool may be retained and used to improve models, so check the provider's terms and ask about removal quickly.
Your first actions are to stop further sharing, preserve evidence, notify leadership and legal counsel, and contact the AI tool provider to request data removal.
Whether customers must be notified depends on your industry, the type of data exposed, your contracts and applicable privacy rules, not only on whether the incident is called a breach.
A clear data-sharing policy and employee training reduce accidental exposure and show regulators and customers that you take care with their information.
What Happens When Data Is Pasted Into an AI Tool?
Pasted text travels to the AI provider's servers and is stored with the account's chat history. Depending on the provider's terms and the plan, it may also be reviewed or used to improve future models. Either way, the data is now outside your control, so the first job is to learn which tool and which plan were involved.
Where your data goes
Free personal accounts, business plans and enterprise agreements can each handle data differently. That makes the specific tool and plan more important than the brand name. A business plan may promise that content is not used for training, while a personal account may carry different terms, so read the provider's current privacy terms instead of relying on memory. It also helps to place the incident inside a broader security program. NIST published The NIST Cybersecurity Framework (CSF) 2.0 as a government reference for managing cybersecurity risk, and it is a reasonable yardstick for how your business handles events like this one.
How long it stays there
Retention varies by provider, plan and settings. Some tools keep chat history until the user deletes it, and deleted chats may remain in provider systems for a period set by the provider's policy. Because those windows can be limited, speed matters, which is why the removal request sits in the first hour. Look up the retention period on the provider's privacy page and note it in your incident record.
Immediate Steps to Take in the First Hour
Three things matter in the first hour, in this order. None of them requires a lawyer or an IT consultant, though you will bring in both soon.
Stop the bleeding
Ask the employee to stop putting work data into that tool and to leave the conversation exactly as it is for now. Deleting it feels like the obvious fix, but it removes your best evidence of what was shared. If passwords, access keys or login details were pasted, change them immediately. If the account is shared, or others may have pasted similar content, ask them to pause too.
Document everything
Capture the facts while memory is fresh. Take screenshots of the conversation, save the conversation link if the tool offers one, and write down the tool name, the account email used, whether the plan was free or paid, the date and time, and what was pasted. Then ask the employee, without blame, why they did it. The answer usually points to a gap in your process, such as a missing approved tool or an unclear rule.
Alert your team
Tell the employee's manager and your own leadership right away. Then give the wider staff only what they need: a short, factual reminder not to paste client or employee data into AI tools until a policy is in place. Leave the individual's name out of it. A plain note from you leaves less room for guessing than office chatter will.
A quick call can change everything.
Notify the Right People and Document the Incident
A coordinated response depends on the right people hearing the news in a sensible order, with the facts written down as you go. Without that order, several people end up making separate decisions on partial information.
Internal escalation order
Work through these in sequence:
Your top decision-maker, such as the owner, executive director or managing partner.
Legal counsel, or whoever advises you on privacy and contracts.
Your IT provider or internal IT contact, who can check whether other accounts or devices are involved.
The person who owns the affected client relationships, once leadership and counsel agree on what can be said.
The order follows the work each person does. Leadership decides how seriously to treat the incident, counsel shapes what is said and written, IT supplies technical facts, and the relationship owner speaks to customers last. Your data breach response for an AI tool like ChatGPT works the same way as for any other incident.
What to document and why
Keep one running incident log, separate from the evidence you captured in the first hour. Record the timeline, who was told and when, what the provider replied, and what was decided and by whom. Stick to facts instead of conclusions, since counsel, a customer or a regulator may read it later. A good log lets you answer "what did you do, and when?" from a document instead of from memory.
Assess What Data Was Exposed and Evaluate Your Risk
Not every paste carries the same weight. A paragraph of public marketing copy and a spreadsheet of client account numbers call for very different responses. Employee data ai tool exposure deserves the same sorting, since HR records and payroll details are as sensitive as customer files.
Data type matters
Start by listing exactly what was pasted, field by field, and whether it identifies a person. Generic or already public text is low risk. Internal plans, client names and work emails sit in the middle. Social Security numbers, financial account details, health information, login credentials and HR records are high risk, and they should move the incident to the front of leadership's agenda. If you cannot tell what was in the paste, treat it as the more sensitive case until the screenshots show otherwise.
Regulatory triggers
Some businesses carry legal duties that others do not. According to the FTC, the Safeguards Rule applies to financial institutions under its jurisdiction, a category broader than most people expect, with examples including mortgage lenders, tax preparation firms and investment advisors not required to register with the SEC. The FTC says it amended the rule in 2023 to require covered entities to report certain data breaches and security incidents, and those requirements took effect in May 2024. The agency's guide, FTC Safeguards Rule: What Your Business Needs to Know, also explains that what matters is the activity your business performs, not how you label yourself. Have counsel check the rule's text to see whether your business is covered and what counts as reportable.
Contact the AI Tool Provider and Request Data Removal
Look in the provider's help center or account settings for a privacy or data deletion request form. Business plans may route this through an account administrator or representative. Send the request in writing so you have a record, and include the following:
The account email and plan type.
The date, time and conversation link.
A short description of the kind of data involved, without pasting it again.
A request to delete the conversation and any associated data.
A question about whether the content was used or queued for model training, and whether it can be excluded.
A request for written confirmation and an expected timeline.
Save the confirmation in your incident log. If the provider does not respond within the time its own policy states, follow up and record that too. A deletion confirmation lowers the risk but does not undo the original exposure, so keep the risk assessment moving while you wait.
Create a Data-Sharing Policy to Prevent It Next Time
A policy that fits on one page is more likely to be read and followed. Start with visibility, because you cannot set rules for tools you do not know about. Our post Shadow AI: How to Find the AI Tools covers how to find the AI tools staff use without approval, which is the starting point for any rule set. For a small business, an AI tool data policy is the clearest way to prevent employee data sharing with AI tools in the first place.
What the policy must cover
Keep it to the decisions your team actually faces:
Which AI tools and plans are approved for work.
Which data may never go into any AI tool, such as client personal information, financial details, health information, credentials and employee records.
What can be shared with care, such as text with names and identifying details removed.
Who to ask before trying a new tool.
What to do after a mistake: report it immediately, with no blame for speaking up.
How to roll it out
Walk through the policy in a short team meeting instead of only sending an email, and ask each person to acknowledge it in writing. Revisit it whenever tools or plans change. The reporting rule matters most, because the sooner someone admits a mistake, the more of that first hour you keep. Giving people an approved tool also removes much of the reason to reach for a personal account. Written rules and training records show regulators and customers that you made a reasonable effort to protect their information.
Mistakes happen—what matters is speaking up right away.
When to Notify Your Customers
A data exposure means information reached a place or person it should not have. A data breach is the term that laws and contracts apply to incidents meeting their specific definitions, and those definitions differ. Whether an employee's paste into an AI tool meets one is a question for your counsel, not something to settle on a hunch.
Several conditions point toward notification: the data is sensitive and identifies individuals, your industry carries reporting duties like the Safeguards Rule discussed above, your client contracts set notice deadlines, or the privacy laws where your customers live require it. That last point matters for Bay Area and Denver organizations that serve customers in other states, because location of the customer, not only of your office, can decide which rules apply. If none of these conditions applies, notification may be optional, though some businesses choose to tell customers anyway.
Bring counsel your incident log, your risk assessment and the provider's confirmation, then record the decision either way. Looking ahead, teams weighing a vetted, company-managed AI setup against ad-hoc tool adoption may find The Real Cost of Building Internal AI Tools useful context.
About 24hourtek
24hourtek, Inc is a forward thinking managed service provider that offers ongoing IT support and strategic guidance to businesses. We meet with our clients at least once a month to review strategy, security posture, and provide guidance on future-proofing your IT.