AI Data Privacy & Safety Checklist
This content was developed with AI assistance and is regularly reviewed for accuracy.
Every prompt you send to an AI tool is data leaving your control. Whether you are drafting an email, analyzing a spreadsheet, or brainstorming strategy, the text you type is processed — and sometimes stored — on servers you do not own. This checklist gives you a practical, day-to-day reference for using AI safely at work, protecting your organization, your customers, and yourself.
Why This Matters
AI tools are extraordinarily useful, but they introduce a category of risk that most workplace policies have not yet caught up with. When you paste content into a chat window, that content may be retained for model training, reviewed by human evaluators, or exposed through a data breach.
Real-world incidents worth knowing about:
- In 2023, Samsung engineers inadvertently leaked proprietary semiconductor source code by pasting it into ChatGPT for debugging assistance. The company subsequently banned generative AI tools internally.
- A 2024 study found that over 10% of enterprise employees had pasted sensitive data — including customer records and API keys — into consumer-tier AI tools at least once.
- Multiple organizations have discovered confidential strategy documents surfacing in AI-generated outputs after employees used consumer accounts for planning work.
These are not hypothetical scenarios. They are the reason every organization needs clear, actionable guidelines — and every individual needs the habit of pausing before pasting.
The Checklist
Use this section as a quick reference. Print it, bookmark it, or share it with your team. Each phase covers a different stage of your interaction with AI tools.
Before You Start
The most common privacy mistakes happen before a single prompt is sent — using the wrong tool, the wrong account, or the wrong assumption about data handling.
- Review your organization's AI usage policy — Know which tools are approved and for what purposes. If no policy exists, flag this with your manager or IT team.
- Understand the platform's data retention policy — Check whether your inputs are stored, for how long, and whether they are used for training.
- Confirm the tool is approved for your use case — An approved tool for marketing copy may not be approved for financial analysis.
- Identify the sensitivity level of your data — Classify what you plan to share as public, internal, confidential, or restricted before you begin.
During Use
Build the habit of stripping names, account numbers, and identifiers before pasting anything into an AI tool. It takes thirty seconds and eliminates the most common category of data exposure.
- Never paste passwords, API keys, or credentials — Treat every AI input field as a public forum. Secrets entered here should be rotated immediately.
- Remove personally identifiable information (PII) — Names, email addresses, phone numbers, and Social Security numbers should never appear in prompts.
- Anonymize customer and employee data before sharing — Replace real names with placeholders like "Customer A" or "Employee 1." Remove dates of birth, addresses, and account numbers.
- Don't share proprietary source code without approval — Code pasted into consumer AI tools may be used for training and could surface in other users' outputs.
- Verify you're using the right account — Enterprise accounts typically have stronger privacy protections than personal accounts. Double-check which one you are logged into.
After Use
- Review AI output for accuracy before sharing — AI can fabricate facts, cite nonexistent sources, and produce plausible-sounding errors. Verify before forwarding.
- Don't assume AI output is confidential — Anything the AI generates may be logged, reviewed, or used in future training unless your agreement explicitly prohibits it.
- Document AI-assisted decisions for audit trails — Note when and how AI contributed to a decision, especially in regulated industries.
- Report any data exposure incidents immediately — If you realize you shared something you should not have, follow your incident response process without delay.
What Never to Share with AI
Some categories of data should never enter an AI prompt under any circumstances, regardless of which platform you are using or what tier of service you have. Understanding why each category is risky helps build lasting judgment, not just rote compliance.
Credentials and secrets — Passwords, API keys, tokens, and certificates. Once shared, these can be extracted from training data or logs. If you accidentally paste a credential, rotate it immediately.
Customer PII — Names, addresses, phone numbers, email addresses, government-issued IDs. Sharing this data may violate your customer agreements and privacy regulations simultaneously.
Raw financial records — Bank statements, transaction logs, payroll data, and revenue figures. These contain both PII and competitively sensitive information. Use aggregated or anonymized summaries instead.
Medical and health data — Patient records, diagnoses, treatment plans, and insurance information. Sharing this outside approved systems can trigger HIPAA violations with penalties reaching millions of dollars.
Legal documents under privilege — Attorney-client communications, litigation strategy, and settlement terms. Sharing privileged documents with a third-party AI service can waive legal privilege entirely.
Proprietary algorithms and trade secrets — Source code, model architectures, manufacturing processes, and formulas. Once in a training dataset, your competitive advantage may appear in a competitor's AI output.
Internal strategic documents — Board presentations, M&A plans, unreleased product roadmaps, and pricing strategies. Premature disclosure can trigger securities violations or competitive harm.
Platform Data Policies (2026)
Understanding how major AI platforms handle your data is essential for making informed decisions about which tools to use for which tasks. The table below summarizes current policies, but always verify directly with the provider — policies change frequently.
| Platform | Data Used for Training? | Enterprise Option | Data Retention | SOC 2 | HIPAA Option |
|---|---|---|---|---|---|
| ChatGPT (OpenAI) | Consumer: Yes (opt-out available). Enterprise: No | ChatGPT Enterprise, Team | 30 days (Enterprise) | Type II | Yes (Enterprise) |
| Claude (Anthropic) | Consumer: No by default. Free: may use | Claude for Business, Enterprise | 90 days (consumer), customizable (enterprise) | Type II | Yes (Enterprise) |
| Gemini (Google) | Consumer: Yes. Workspace: No | Google Workspace add-on | 18 months (consumer), customizable | Type II | Yes (Workspace) |
| Microsoft Copilot | Consumer: Yes. Commercial: No | Microsoft 365 Copilot | Aligned with M365 retention | Type II | Yes (with M365) |
| Perplexity | Consumer: Yes | Perplexity Enterprise Pro | Not publicly specified | Pending | No |
Consumer and enterprise tiers of the same platform often have fundamentally different data handling practices. A "free" account at ChatGPT or Gemini may use your inputs for training; the paid enterprise version of the same product typically does not. Never assume your personal experience with a tool reflects how the enterprise version works — or vice versa.
Enterprise Security Requirements
When evaluating AI tools for organizational deployment, these capabilities separate production-ready platforms from consumer toys. Any enterprise AI deployment should include the following.
SSO/SAML integration — Employees should authenticate through your existing identity provider. This eliminates shared passwords and enables centralized access control.
Data residency controls — For organizations subject to data sovereignty laws, the ability to specify where data is processed and stored (e.g., EU-only, US-only) is non-negotiable.
Audit logging — Every interaction should be logged with timestamps, user identifiers, and content summaries. This supports compliance audits and incident investigations.
Role-based access control — Not every employee needs access to every AI capability. Restrict sensitive use cases (e.g., code generation, data analysis) to appropriate roles.
Data processing agreements (DPAs) — A signed DPA clarifies the provider's obligations under privacy regulations. If a vendor will not sign one, that is a disqualifying signal.
SOC 2 Type II compliance — This attestation confirms that the provider's security controls have been independently audited and found effective over time, not just at a single point.
Custom data retention — The ability to set your own retention periods — including zero-day deletion — ensures alignment with your data governance policies.
Regulatory Quick Reference
AI use intersects with multiple regulatory frameworks. This table provides a starting point; consult your legal team for guidance specific to your industry and jurisdiction.
| Regulation | Scope | AI-Relevant Requirements | Penalties |
|---|---|---|---|
| GDPR | EU residents' data | Lawful basis for processing, right to explanation of automated decisions, data minimization, DPIAs for high-risk AI | Up to 4% of global annual revenue or 20M EUR |
| CCPA/CPRA | California residents' data | Right to know about automated decision-making, opt-out of data sales, data deletion rights | $2,500 per violation, $7,500 per intentional violation |
| HIPAA | Protected health information (US) | Requires BAAs with AI vendors, minimum necessary standard, access controls, audit trails | Up to $2.1M per violation category per year |
| EU AI Act | AI systems in the EU market | Risk classification (unacceptable, high, limited, minimal), transparency obligations, conformity assessments for high-risk systems | Up to 35M EUR or 7% of global turnover |
GDPR remains the most far-reaching regulation for AI use. If your organization processes data from EU residents, every AI tool you use must have a lawful basis for processing, and any automated decisions that significantly affect individuals must be explainable.
CCPA/CPRA gives California residents the right to know when automated decision-making is used and to opt out of having their data sold — including to AI training datasets.
HIPAA requires a Business Associate Agreement (BAA) with any AI vendor that processes protected health information. Using a consumer AI tool for anything involving patient data is a violation, full stop.
The EU AI Act is now well into its phased rollout — the bans on unacceptable-risk systems took effect in February 2025. High-risk system obligations were originally slated to begin August 2026, but the EU's "Digital Omnibus" agreement, finalized in June 2026, delayed most standalone high-risk requirements to December 2027 (with embedded-product obligations into August 2028). High-risk systems — including those used in employment, credit, and law enforcement — will still face mandatory conformity assessments and ongoing monitoring requirements once their deadline lands; confirm the current status with your legal team before finalizing a compliance timeline.
Team Best Practices
Individual habits matter, but organizational systems matter more. These practices help teams adopt AI safely and consistently.
Create an approved tools list. Maintain a living document of AI tools that have been vetted for security, compliance, and data handling. Make it easy to find — if people have to search for it, they will skip the step.
Establish data classification guidelines. Define what counts as public, internal, confidential, and restricted data. Map each classification to specific AI usage rules (e.g., "confidential data may only be used with enterprise-tier tools that have signed DPAs").
Provide training on safe AI use. One-time training is insufficient. Run quarterly refreshers, share incident summaries (anonymized), and celebrate teams that catch potential exposures before they happen.
Designate an AI safety point of contact. Give people a specific person to ask when they are unsure whether a use case is safe. Ambiguity leads to risk; a named contact leads to questions.
Conduct periodic access reviews. Audit who has access to which AI tools and whether that access is still appropriate. Remove accounts for departed employees and reduce access for role changes.
Share anonymization techniques. Teach your team practical methods for stripping sensitive data from prompts — find-and-replace patterns, template prompts with placeholder variables, and data masking tools.
For a deeper dive into building organizational AI policies, see AI Governance Frameworks.
Incident Response
If sensitive data has been shared with an AI tool, act quickly and methodically. Speed matters — some platforms process and distribute data rapidly, and early intervention can limit exposure.
- Stop sharing additional data immediately. Do not continue the conversation or attempt to "fix" the exposure by providing more context.
- Document what was shared and when. Screenshot the conversation, note the timestamp, the platform, the account used, and the specific data exposed. Preserve this record.
- Check the platform's data deletion process. Most enterprise AI platforms offer data deletion requests. Consumer platforms may offer conversation deletion, but this does not guarantee removal from training data.
- Notify your security or privacy team. Follow your organization's incident reporting procedures. If none exist, notify your manager and IT security directly.
- File a support request with the AI provider. Request explicit confirmation that the data has been deleted from all systems, including training pipelines and backups.
- Review and update procedures. Every incident is a learning opportunity. Update your team's guidelines to prevent recurrence and share the lesson (anonymized) broadly.
If your incident response process punishes people for reporting mistakes, people will stop reporting. Design your process to encourage transparency and focus on systemic improvements.
Key Takeaways
- Every AI prompt is data leaving your control. Treat input fields with the same caution you would treat sending an email to an external contact.
- Consumer and enterprise tiers are fundamentally different. Never assume a personal account has the same protections as your company's enterprise agreement.
- Anonymize by default. Strip names, identifiers, and sensitive details before pasting anything. It takes seconds and prevents the most common category of exposure.
- Know your regulations. GDPR, HIPAA, CCPA, and the EU AI Act all have specific requirements for AI use. Ignorance is not a defense.
- Build organizational systems, not just individual habits. Approved tools lists, data classification guidelines, and named safety contacts create durable protection.
- When in doubt, don't paste. If you are unsure whether something is safe to share, stop and ask your AI safety point of contact.
Next Steps
- Data Privacy & Security for foundational concepts on how AI handles your data
- AI Governance Frameworks for building organizational AI policies
- AI Policy Template for creating your organization's AI acceptable use policy
- Automation Best Practices for secure automation implementation