GPT-6-Luna

FDA's public AI-enabled device list contains hundreds of authorized products, and radiology represents its largest specialty. Signals imaging teams need device-level oversight for the clinical AI category with the densest authorized product base.

Judge · The FDA confirms over 1,600 AI-enabled devices, with radiology accounting for a significant majority (76.5% - 76%) of authorizations. This trend is well-documented and current.

Writing · Concrete actor (FDA, radiology), event (authorized products), and specific observation.

FDA-authorized autonomous retinal screening identifies more-than-mild diabetic retinopathy from retinal images without clinician interpretation. Indicates primary-care pathways can place diagnostic decisions at the software boundary, requiring clear escalation ownership.

Judge · FDA-authorized autonomous AI for diabetic retinopathy exists and is used in primary care, shifting diagnostic decision-making to software. [accessdata.fda.gov](https://www.accessdata.fda.gov/cdrh_docs/reviews/DEN180001.pdf), [doi.org](https://doi.org/10.1038/s41746-018-0040-6), [diabetesjournals.org](https://diabetesjournals.org/clinical/article/42/1/142/153640/Clinical-Implementation-of-Autonomous-Artificial), [pmc.ncbi.nlm.nih.gov](https://pmc.ncbi.nlm.nih.gov/articles/PMC8120059/), [doi.org](https://doi.org/10.1001/jamanetworkopen.2025.38114)

Writing · Concrete actor (FDA, software) and event (screening, identifies). Lacks a temporal anchor.

Health systems deploy ambient AI scribes that convert clinician-patient conversations into draft notes for clinician review. Signals documentation automation changes clinical accountability where generated text enters the medical record.

Judge · Health systems are deploying ambient AI for draft notes, but the shift in clinical accountability due to AI-generated text is a recognized, ongoing challenge.

Writing · Concrete actor (hospital systems) and event (deploy AI). Lacks specific company/product names, dates, or numbers.

Published clinical evaluations document inaccurate or unsupported answers from large language models on medical questions. Indicates clinical teams need explicit limits and escalation paths before model outputs inform diagnosis or treatment.

Judge · Multiple studies confirm LLMs provide inaccurate/unsupported clinical outputs. This necessitates clear limits and escalation paths for AI in healthcare.

Writing · Concrete actor (clinical evaluations), event (inaccurate answers), and a clear implication. Lacks a specific model or date.

The EU AI Act designates certain AI medical-device products requiring third-party conformity assessment as high-risk systems. Signals hospital networks need coordinated conformity and risk controls across affected clinical products.

Judge · The EU AI Act classifies medical devices requiring third-party conformity assessment as high-risk. This necessitates coordinated controls for hospital networks using such AI products.

Writing · Concrete actor, event, and anchor, but lacks a specific product/filing. Contains some generic forecast.

FDA's December 2024 final guidance describes predetermined change control plans for planned modifications to AI-enabled device software. Indicates manufacturers and health systems need defined evidence and approval processes for model updates.

Judge · FDA's December 2024 final guidance details PCCPs for AI-enabled device software, requiring evidence and approval for model updates. This will impact manufacturers and health systems.

Writing · Concrete actor (FDA), event (finalizing framework), and measurable shift (new regulatory pathway) are present. Lacks a temporal anchor.

ONC's HTI-1 rule requires certified health IT developers to disclose source attributes for predictive decision support interventions. Signals hospitals can examine more information when assessing algorithms embedded in certified systems.

Judge · The ONC HTI-1 rule mandates transparency for predictive decision support interventions (DSI) in certified health IT, effective January 1, 2025. This allows hospitals to access more algorithm information.

Writing · Names ONC and HTI-1 rule, details 'transparency and risk-management requirements,' and 'model documentation, evaluation evidence, and risk-management information.' Strong specifics. Present tense.

The EU's European Health Data Space Regulation enters into force in 2025 and phases secondary-use access to electronic health data. Signals hospital networks need governed data access for AI development across borders and care settings.

Judge · The EHDS regulation enters force March 2025, enabling phased secondary use of health data across the EU for AI training and other purposes under strict governance. This is well-documented.

Writing · Names a concrete regulation and its purpose. Lacks a quantitative/temporal anchor or a specific actor impacted.

Health systems deploy ambient AI scribes that convert clinician-patient conversations into draft notes for clinician review. Signals operations teams must define review, correction, and record-signoff steps within clinical workflows.

Judge · Health systems are deploying ambient AI for draft notes, but the shift in clinical accountability due to AI-generated text is a recognized, ongoing challenge.

Writing · Concrete actor and event, but lacks quantitative/temporal anchor and uses some future-tense implications.

EHR vendors add generative AI features for drafting notes, summarizing charts, and preparing patient-message replies. Indicates health networks must assess these tools within existing access, configuration, and quality-control processes.

Judge · EHR vendors are integrating generative AI for drafting, summarizing, and replying. Health systems must evaluate these tools within existing frameworks.

Writing · Concrete actors (EHR vendors) and actions (drafting notes) are named. No specific product or timeline.

Hospitals obtain AI functions through EHR, imaging, cloud, and specialist vendors, with updates governed by separate service arrangements. Signals procurement and security teams need visibility into vendor dependencies and model changes.

Judge · The need for hospitals to gain visibility into third-party AI supply chains and model changes is well-documented by multiple industry sources. [healthsectorcouncil.org](https://healthsectorcouncil.org/wp-content/uploads/2026/04/AI-Third-Party-Risk-Guide.pdf) [latticehealthai.com](https://www.latticehealthai.com/) [censinet.com](https://censinet.com/perspectives/hidden-third-parties-healthcare-ai) [ai.nejm.org](https://ai.nejm.org/doi/full/10.1056/AIp2500985) [healthaipolicy.org](https://www.healthaipolicy.org/policies/Third-Party%20AI%20Risk%20and%20Supply%20Chain%20Transparency%20Guide)

Writing · Concrete actors (hospitals, EHR, cloud vendors) and a clear problem (visibility needed). Lacks specific named examples or quantifiers.

The Coalition for Health AI's assurance guidance calls for local validation, monitoring, and documentation of healthcare AI models. Indicates health networks need assigned teams and repeatable processes to verify performance in local settings.

Judge · CHAI's guidance explicitly recommends local validation, ongoing monitoring, and documentation for healthcare AI models. It highlights the need for dedicated resources and processes.

Writing · Names actor and concrete event, but lacks quantitative/temporal anchor and active voice for a higher score.

A 2023 Pew Research Center survey finds most US adults uncomfortable with providers relying on AI for diagnosis or treatment recommendations. Signals patient acceptance is relevant to clinical deployment and communication choices.

Judge · Multiple reputable sources, including Pew Research itself, confirm the survey findings and date. Patient acceptance is directly linked to AI deployment.

Writing · Concrete actor, event, and temporal anchor. 'Most' is slightly vague but survey-based. Active voice.

Health systems offer ambient documentation tools that capture visits and generate draft notes, with patient notice and opt-out processes varying by site. Indicates consent and recording explanations shape patient expectations during clinical encounters.

Judge · Health systems use ambient AI documentation with varied consent processes. Patient understanding of recording and data use is crucial. Opt-out options exist but consistency varies.

Writing · No concrete actor, event, or specific anchor. 'Varying by site' is vague. Active voice is good.

The EU AI Act sets disclosure duties for people interacting directly with certain AI systems, subject to stated exceptions. Signals patient-facing services need clear distinctions between automated assistance and human communication.

Judge · The EU AI Act (Article 50) mandates disclosure when people interact directly with AI, with specific exceptions. This applies from August 2026, signaling a clear need for patient-facing services to distinguish AI from human interaction.

Writing · Names the EU AI Act and 'patient-facing services' but lacks a specific event, filing, or quantitative anchor.

HIPAA permits covered entities to use properly de-identified health information without individual authorization, while GDPR classifies health data as special-category personal data. Signals explanations of data use matter to patient confidence in AI development.

Judge · HIPAA allows de-identified data use; GDPR classifies health data as 'special category.' Transparency in data use is crucial for patient trust in AI.

Writing · No concrete actor, event, or quantitative anchor. Uses present tense appropriately.