Medical AI Compliance Glossary

Evaluation and minimization of risks.

A security-relevant chronological record providing documentary evidence of the sequence of activities that have affected a specific operation, procedure, or event.

The deceptive design practice where an AI system is anthropomorphized to mimic a licensed healthcare professional (e.g., wearing white coats).

The legal mandate that AI systems must inform users they are interacting with artificial intelligence at the start of every interaction.

A system design requiring human review of AI output before it reaches the patient.

Giving a false account of the nature of something (e.g., AI appearing human).

Requirement that AI notices remain visible throughout an interaction.

Credentials like M.D. or R.N. after a name.

Prohibition on AI avatars wearing clinical attire.

Designing interfaces that meet legal standards (e.g., font size).

Principle that AI should be understandable and disclosed.

Remote delivery of healthcare.

Potential responsibility for payment of damages or other court-enforced remedies in a lawsuit.

Any profession subject to licensure under the Healing Arts division.

The state permit to practice medicine.

Information relating to a patient's health status.

California Assembly Bill 2013 requires developers of generative AI systems to provide high-level summaries of the datasets used to train their models. This includes information on the categories of data, whether it contains PII, and the time period it covers.

California Assembly Bill 3030 is a focused regulation targeting the use of Generative AI (GenAI) in healthcare communications. It mandates that if a GenAI tool is used to communicate clinical information to a patient, a human healthcare provider must review and approve that communication before it is sent.

California Assembly Bill 489, formally known as the 'Artificial Intelligence Transparency Act for Healthcare,' is a state law enacted to prevent the deceptive use of AI in medical settings. It mandates that any automated system interacting with a patient must clearly and conspicuously disclose strictly that it is not a human.

Federal law protecting sensitive patient health information (PHI).

The process of proving that an AI model works as intended in a real-world clinical setting.

Permission obtained from a patient with full knowledge of the risks.

Patient's right to be informed when interacting with AI.

Protected Health Information under HIPAA.

Government bodies like the Medical Board of California empowered to uphold AI laws.

State agency licensing physicians and enforcing medical laws.

Regulatory body for DOs in California.

Experience, Expertise, Authoritativeness, and Trustworthiness. Google's quality framework.

Your Money or Your Life content classification.

Systematic and repeatable errors in a computer system that create unfair outcomes, such as privileging one arbitrary group of users over others.

Any computational system that performs tasks indicative of human intelligence, broadly defined in legislation to distinguish valid tools from misleading automation.

Health IT tools providing clinical knowledge and patient-related information to enhance health care.

Documentation of the origin and history of the data used to train an AI model.

The connection between AI systems and electronic health record software.

AI models that can create new content (text, images) rather than just analyzing data.

When an AI produces confident but factually incorrect information.

AI trained on vast text data to generate human-like language (e.g., GPT-4).

The dataset used to teach an AI model.

AI interacting directly with patients (chatbots, voice assistants).

Software performing medical purposes without hardware.

AI assessing symptoms to recommend care urgency.

Content generated by AI mimicking human creation.

AI help for administrative/triage tasks.

AI interacting via spoken language.