How One Spreadsheet Uploaded to a Public AI Tool Became a HIPAA Breach at Johns Hopkins

During an internal review on September 21, 2026, Johns Hopkins Bayview Medical Center in Baltimore discovered that an employee mistakenly uploaded a spreadsheet containing patient information to an unapproved public artificial intelligence (AI) tool. This type of unauthorized use is referred to as shadow AI.

The spreadsheet contained information about patients admitted or treated between May and September 2026, potentially including names, treatment dates, medical record and account numbers, medications, and provider names. It did not include Social Security numbers or financial account information. 

Following the discovery, the hospital launched an investigation, had the employee delete the uploaded information from the tool’s history, and requested additional deletion by the AI provider. On October 22, Johns Hopkins began mailing notification letters to 3,182 potentially affected patients. The hospital is also reinforcing its AI policies and workforce training. 

The incident illustrates the growing risks of shadow AI, a cybersecurity and HIPAA compliance concern for healthcare organizations. 

What Is Shadow AI in Healthcare? 

Shadow AI occurs when employees use artificial intelligence tools, accounts, plugins, or integrations for work without their organization’s approval or oversight. Examples include using a free chatbot to summarize meeting notes, a browser extension to draft patient emails, or a personal AI account to organize spreadsheets. 

While most employees likely have good intentions to work more efficiently, these tools may process sensitive information outside the organization’s approved security environment. When AI use is invisible to IT and security teams, the information being processed may be invisible as well. 

The Johns Hopkins incident is particularly noteworthy because the institution had already published AI usage guidelines in May 2025, instructing employees not to enter clinical or other proprietary information into unapproved AI tools. 

How Does Shadow AI Put HIPAA Compliance at Risk? 

Under HIPAA, healthcare organizations must protect patient information by implementing reasonable and appropriate administrative, physical, and technical safeguards. The HIPAA Security Rule requires organizations to assess risks to electronic protected health information (ePHI) and implement safeguards appropriate to their size, complexity, capabilities, and identified risks. 

Shadow AI Can Introduce Several Significant Risks

• Unauthorized PHI disclosures: Employees may inadvertently expose patient information by uploading it to public AI applications. 
• Missing business associate agreements (BAAs): AI providers handling PHI on behalf of healthcare organizations generally require a HIPAA-compliant BAA. 
• Uncertain data retention: Organizations may not know whether information is stored, reused, or permanently deleted by an AI provider. 
• Inadequate security safeguards: Unapproved applications may lack verified encryption, access controls, monitoring, and other necessary protections. 
• Limited visibility and accountability: IT and compliance teams cannot effectively manage risks involving applications they do not know employees are using. 

According to HHS guidance on cloud services and HIPAA, healthcare organizations may use third-party cloud platforms to process ePHI when they have required BAAs and comply with applicable HIPAA requirements. 

However, simply obtaining a BAA does not guarantee compliance. Organizations must also conduct appropriate risk analyses, implement reasonable and appropriate safeguards, and ensure that PHI is used and disclosed only as permitted. 

HIPAA does not mandate specific technologies, such as DLP software, to prevent shadow AI incidents. Instead, organizations must evaluate their risks, select appropriate security measures, and document their decisions in accordance with the HIPAA Security Rule. 

6 Steps to Manage Shadow AI and Protect Patient Data 

Healthcare organizations need more than policies prohibiting unauthorized AI. They need practical governance programs that make approved, secure technology accessible and easy to use. 

1. Identify AI tools already in use. Start with a judgment-free workforce survey to understand which AI applications employees rely on. Review existing tools, browser extensions, and integrations to identify potential unauthorized use.
2. Include AI in HIPAA security risk assessments. Evaluate how AI applications collect, process, store, and transmit ePHI. Identify potential vulnerabilities, vendor risks, and gaps in existing security controls. The HHS Security Rule risk analysis guidance emphasizes assessing risks across all ePHI an organization creates, receives, maintains, or transmits.
3. Provide approved AI alternatives. Select AI tools that support legitimate workforce needs while meeting privacy and security requirements. Evaluate vendors, establish required BAAs, and verify safeguards before permitting PHI processing.
4. Implement reasonable and appropriate technical safeguards. Based on the organization’s risk assessment, consider protections such as DLP software to detect or block unauthorized PHI uploads, access controls, encryption, monitoring, and audit logging. While HIPAA does not require a specific DLP technology, organizations must implement safeguards that reasonably address identified security risks and meet applicable Security Rule standards.
5. Establish clear policies and employee training. Develop clear guidelines that explain which AI tools are approved, what information employees may enter, and who to contact with questions. A one-page AI use guide, reinforced through regular HIPAA training, can help employees make safer decisions.
6. Monitor AI use and prepare for incidents. Regularly review AI applications and vendor practices. Establish procedures to investigate unauthorized disclosures, document corrective actions, request data deletion, and assess potential HIPAA breach notification obligations.

Under the HIPAA Breach Notification Rule, an impermissible disclosure of unsecured PHI is generally presumed to be a breach unless an applicable exception applies or a documented assessment demonstrates a low probability that the information was compromised. This makes prompt investigation and documentation especially important when shadow AI incidents occur. 

Make Secure AI Use the Easier Choice 

The Johns Hopkins incident demonstrates that even organizations with established AI policies can experience unauthorized disclosures when employees turn to convenient, unapproved tools. 

Banning AI alone is not the answer. Healthcare organizations need visibility into how employees use AI, practical alternatives that meet workforce needs, and reasonable and appropriate safeguards based on their unique security risks. 

The goal is to make responsible, HIPAA-compliant AI use the easiest choice, not an obstacle to getting work done. 

FAQs about the Johns Hopkins Shadow AI Breach