Example scenario
It's 4:45 on a Thursday. A recruiter has a background report open with a flagged record she doesn't quite understand, and a hiring manager is waiting for an answer. She pastes the report into ChatGPT and asks, "Is this disqualifying?" The answer comes back in seconds. So does a problem: she has just sent a consumer report, a Social Security number and a date of birth to an outside AI provider, with no audit trail to show where it went.
The short answer: what HR should never put into ChatGPT is applicant or employee personal data. That includes background reports, Social Security numbers, medical information and adverse action letters, and the same goes for Claude, Gemini, Copilot and any other consumer AI tool. Below are the 11 types of data to keep out, what the law says and what to do instead. For a version you can share with your team, download the free one-page AI and Employee Data Checklist.
11 Types of Data HR Should Never Enter Into ChatGPT
- Direct identifiers (Social Security number, date of birth and driver's license number). One paste sends personal data to an outside AI provider with no deletion guarantee.
- Background screening reports. Criminal records, verifications, credit data and motor vehicle records prepared by a screening company are consumer reports under the FCRA. That classification triggers disclosure, authorization, adverse-action and disposal duties for you, and accuracy and dispute duties for your screening provider.
- Adverse-action materials. Pre-adverse and final adverse-action letters contain regulated information. AI output can't replace the human review a fair, defensible decision needs, including the individualized assessment many state and local fair chance laws require, which KRESS tracks in its State-by-State Compliance Guide.
- Medical and disability data. Prescription disclosures, accommodation requests, diagnoses and workers' comp records fall under the Americans with Disabilities Act (ADA), FMLA confidentiality rules and state medical privacy laws.
- Genetic and family medical information. GINA bars employers from requesting, requiring or purchasing genetic information, including family medical history, and requires any they hold to be kept as a confidential medical record. Pasting it into an AI tool is a disclosure outside those files.
- Protected-class information. Race, sex, pregnancy, age, religion, disability, veteran status, national origin and sexual orientation. A prompt containing this data could become evidence in a discrimination claim.
- Financial and identity documents. Bank details, tax forms, wage records and I-9 documents carry identity-theft risk.
- Investigation and legal data. Complaints, witness interviews, litigation strategy and attorney communications. Uploading these could put attorney-client privilege at risk or compromise an investigation.
- Entire personnel files. HRIS exports, disciplinary histories and performance reviews combine several sensitive categories at once.
- Unredacted resumes or applications. A graduation year signals approximate age, and a school name can indicate race, religion or national origin.
- Applicant disputes. Dispute letters and explanations of criminal records need careful human evaluation and often hold deeply personal context.
The FCRA Problem With Background Reports in ChatGPT
People paste reports into AI for a simple reason: flagged results are hard to read alone. That's a good moment to ask a person, not a chatbot. Your KRESS team can walk you through a record you don't understand, inside the platform where the report belongs.
Pasting a background report into ChatGPT breaks the FCRA compliance chain. It creates an extra copy outside your screening file. If the applicant disputes the report, your screening provider reinvestigates and corrects its file, but the copy in the chat never gets that correction. And it creates discoverable evidence that you handed a regulated employment decision to an opaque system.
Keep reports inside your screening platform, limit access to trained reviewers and use a compliant adverse-action process for each FCRA step. With KRESS, the report stays in the platform and the hiring decision stays with your team.
Don't Let AI Make the Hiring Decision
Never ask an AI tool:
- "Should we hire this person?"
- "Is this criminal record disqualifying?"
- "Rank these applicants by trustworthiness."
- "Infer whether this applicant has a disability."
They invite the system to infer protected characteristics or lean on proxies, creating Title VII, ADA and GINA exposure. Title VII bars selection practices with an unjustified disparate impact, whatever tool produces them, and the Justice Department's ADA guidance warns that employers violate the ADA if hiring technology unfairly screens out a qualified individual with a disability. The ADA's limits on medical inquiries during background checks apply here too.
A sound baseline: a qualified human makes the final call and can explain it without an AI score, your team tests for disparate impact, and applicants can ask for accommodation or correction.
State and Local AI Hiring Laws
A growing number of places regulate AI in hiring. Colorado's first AI Act never took effect: SB 26-189 replaced it in May 2026, before its June 30 date, and the new rules start January 1, 2027. See our post on AI hiring bias compliance for how California, New York City and Illinois approach it. KRESS tracks the rules where you hire in its State-by-State Compliance Guide.
Safer Alternatives: What You Can Do Instead
You don't have to abandon AI. You need clear boundaries.
Unsafe prompt: "Review this applicant's background report and tell me whether we should reject them."
Safer prompt: "Create a neutral checklist a trained HR reviewer can use to check that our job-related criminal-history policy is applied consistently. Do not use or infer protected characteristics, medical information or personal identifiers."
Other safe uses include synthetic applicant profiles for training, templates with redacted examples and de-identified aggregate data. Real employee or applicant data belongs in approved enterprise tools with legal sign-off.
What to Do If Someone Already Uploaded Sensitive Data
Treat it as a potential data incident.
- Stop further uploads immediately.
- Preserve evidence of what was submitted, when, by whom and under which account type.
- Request deletion through the AI provider's deletion process.
- Notify privacy, security, legal and HR leadership.
- Determine whether the data included a consumer report or other FCRA-regulated information, which carries its own handling and disposal duties.
- Assess state breach-notification requirements and document your response.
- Retrain the team and add technical controls to prevent a repeat.
Frequently Asked Questions
Can I paste a background check report into ChatGPT to summarize it?
No. It sends a consumer report to an outside AI provider and leaves a copy your screening provider can't correct if the applicant disputes it.
Is ChatGPT Enterprise safe for HR data?
"Not used for training" helps, but it's not a legal safe harbor. Deployment still needs legal, privacy and security review against your FCRA, ADA and GINA duties.
Can I use ChatGPT to decide if an applicant's criminal record is disqualifying?
No. FCRA adverse-action rules require you to give the applicant a copy of the report and a summary of their rights before taking adverse action based on it, and many state and local fair chance laws call for an individualized assessment by a qualified person. KRESS's State-by-State Compliance Guide tracks the rules where you hire.
What should I do if an employee already put applicant data into ChatGPT?
Treat it as a potential data incident and follow the seven steps above, starting with stopping further uploads and preserving evidence.
You shouldn't have to become an AI privacy lawyer on top of everything else you manage, but one pasted report can leave a copy you can't correct and evidence you can't take back. Download the free AI and Employee Data Checklist and share it with your team before the next report gets pasted. Questions about where your reports go today? Book a screening workflow review.



