Is it safe to share {X} with {Y}?

No. Sharing protected health information with ChatGPT is not safe under normal circumstances. OpenAI's default data practices allow user inputs to be retained and reviewed, meaning PHI entered into ChatGPT can leave your control the moment it is submitted. Without a signed Business Associate Agreement, there is no HIPAA-compliant framework governing how that information is handled or stored.

Why this matters

OpenAI is not a HIPAA-covered entity by default, so PHI shared through the standard platform receives no HIPAA protections.
Inputs may be retained for up to 30 days and used to improve OpenAI's models unless users have opted out through specific enterprise agreements.
Once PHI is transmitted outside a covered system, your organization loses audit control over who accessed it and under what conditions.

For enterprise

Employees using personal or unauthorized ChatGPT accounts to process PHI create direct compliance exposure for their organizations. This behavior can constitute an unauthorized disclosure under HIPAA, triggering breach notification obligations regardless of intent. Standard enterprise IT policies should explicitly prohibit PHI entry into any AI tool not covered by a formal data processing agreement.

Compliances at risk

What counts as PHI?

Protected Health Information (PHI)
Patient identifiers
Medical treatment information
Health insurance details
Clinical records

Why people share PHI with ChatGPT

To summarize patient information
To prepare clinical documentation
To review treatment history
To draft healthcare reports

What actually happens when you paste PHI into ChatGPT

When you paste PHI into ChatGPT, that data is transmitted from your device to external servers operated by the AI provider.

Depending on system configuration and policies, the data may be logged, temporarily stored, or reviewed for safety and quality purposes. Retention can last from days to weeks, and in some cases may extend beyond the immediate session.

Statements such as “we do not train on your data” do not eliminate risks related to retention, logging, or internal access. These controls vary by product and setting, and are not always visible to end users.

From a governance perspective, any non-zero retention window introduces exposure risk when sensitive data is shared without controls, auditability, or enforcement.

Risks of sharing PHI with ChatGPT

Patient privacy violations: Protected health information may be exposed without authorization.
Regulatory penalties: Improper disclosure can violate healthcare privacy regulations such as HIPAA.
Medical identity theft: Health records can be exploited for insurance fraud or identity misuse.

Real incidents

Training data extraction : Researchers extracted memorized personal and sensitive information.
Italy temporarily bans ChatGPT : Regulators raised concerns over GDPR compliance.
Italy fines OpenAI €15M : Privacy regulators found GDPR violations in ChatGPT's data processing.

Is this allowed under policy or law?

Context	Is it safe?
Personal experimentation	No
Business use	No
Regulated industry	Definitely not
With redaction	Rarely

Safer ways to handle PHI

PHI should not be shared with consumer AI tools without controls in place. If AI assistance is required, organizations should use systems that enforce data redaction, access controls, and policy enforcement before data leaves their environment.

Automatically redact sensitive fields before sending data to AI models
Prevent unauthorized data from being entered into external tools
Maintain audit logs and visibility into how data is used
Ensure compliance with frameworks like GDPR, CCPA, and SOC 2

Platforms like Wald are designed to enable safe AI usage by ensuring sensitive data never leaves your control unprotected.

How Wald.ai handles this safely

Wald adds a governance layer to AI usage, helping organizations monitor and control how sensitive data like PHI is shared.

AI DLP

Identifies PHI in context and enables teams to:

Observe AI usage
Detect sensitive data in prompts
Allow, warn, or block actions
Maintain audit logs

LLM Pack

Provides controlled access to multiple AI models (ChatGPT, Claude, Grok, and others) through a single governed environment.

Centralized model access
Policy enforcement
Usage visibility
Auditability

Frequently Asked Questions

Is it safe to share PHI with ChatGPT?

No. PHI should not be shared with ChatGPT. Exposure can create security, privacy, or compliance risks, and once submitted there may be limited control over retention, logging, or downstream processing.

What happens when PHI is entered into ChatGPT?

The data is transmitted to the AI provider's infrastructure for processing. Depending on the service and configuration, it may be temporarily stored, logged, or retained for security and operational purposes.

Can ChatGPT retain PHI after a conversation ends?

ChatGPT providers may temporarily retain prompts and responses for security, abuse monitoring, or operational purposes. Depending on the platform and settings, PHI may remain stored beyond the immediate session. In some cases, submitted data may be retained for up to 30 days before deletion. Organizations should assume that any sensitive information shared with AI systems could persist beyond the active conversation.

Does ChatGPT train on PHI?

Some AI providers allow organizations to disable training on submitted data, while others may use interactions to improve services. Even when training is disabled, PHI may still be processed, logged, or retained according to provider policies.

What happens if PHI is accidentally shared with ChatGPT?

Once submitted, organizations may have limited visibility into how the information is retained, processed, or accessed. The appropriate response depends on the sensitivity of the data, internal policies, and incident response procedures.

Why do traditional DLP solutions struggle to identify PHI in AI prompts?

Traditional DLP tools rely heavily on pattern matching and predefined rules. AI prompts often contain fragmented, transformed, or contextual information that can be difficult to classify accurately. Context-aware AI DLP solutions can evaluate surrounding context to better distinguish between similar data types and reduce false positives and false negatives.

Secure Your Employee Conversations with AI Assistants

Book A Demo