MEDICAL AI SAFETY / RED TEAMING / CLINICAL EVALUATION
A plausible answer can still be clinically unsafe.
Physician-led testing of model behavior, safety boundaries, and reproducible failures in clinical contexts.
Featured evidence · Medication safety
When recognizing a risk does not preserve the safety boundary.
In 80 fresh-chat conversations, the study examined whether Claude Opus 5 kept a medication-safety boundary intact under changes in role and authority framing.
Read the observed outcomes alongside the experimental conditions, coding rules, and limitations.
Read the medication-safety study →
Author-posted manuscript · Not peer reviewed · August 2026 experiments
THE EVALUATION QUESTION
What changes when the clinical context changes?
Accuracy is one part of an evaluation. The decision to help, refuse, escalate, or qualify an answer also needs to be tested.
01 / RESPONSE
Unsafe compliance
Does the answer cross a clinically meaningful safety boundary?
02 / CONTEXT
Role-sensitive safeguards
Does a claimed credential or a change in context alter the response?
03 / RELIABILITY
Reproducibility
Does the behavior recur under matched, documented conditions?
COMMISSIONED WORK
A defined question. A traceable deliverable.
EVALUATE
Clinical AI safety review
A scoped assessment of outputs, clinical context, and safety-relevant failure patterns.
CHARACTERIZE
Medical red teaming
Matched conditions and repeat testing to characterize when a boundary changes.
BUILD
Benchmarks and simulations
Physician-authored scenarios, scoring rubrics, and interactive educational pathways.
INTERACTIVE SYSTEMS
See the interaction, then discuss the application.
CRUCIBLE · PUBLIC DEMOS
Clinical decisions before feedback
Staged case challenges with diagnostic commitment and configurable confidence and timing layers.
CASCADE · SELECTED PUBLIC COMPONENTS
Advanced therapy education
Mechanism-rich learning journeys. Public components demonstrate the format; full modules are scoped by request.
QUINTUPLE-BOARD-CERTIFIED PHYSICIAN
Why Olga Goodman, MD
Physician-founder & Clinical AI Evaluation Architect. Five U.S. board certifications, active rheumatology practice, and original evaluation-system development bring clinical breadth and practical accountability to AI review.
CLINICAL BREADTH
Reason across specialties
Rheumatology, internal medicine, pediatrics, pain management, and regenerative/aesthetic medicine inform review of the interactions between diagnosis, comorbidity, treatment, and procedural risk.
DIRECT CLINICAL EXPERIENCE
Read the evidence behind the answer
Personal interpretation of radiographs and musculoskeletal ultrasound, alongside ultrasound-guided procedures, informs scrutiny of unsupported inferences and mismatches between clinical and imaging evidence.
EVALUATION ARCHITECTURE
Make clinical judgment testable
Physician-authored reference answers, safety rubrics, severity anchors, and repeatable scenarios translate clinical concerns into findings that developers can investigate and retest.
Selected recognition: ACR Annual Image Competition — Grand Prize of the year, 2014; InterACT Annual Interdisciplinary Autoimmune Summit — Grand Prize, 2022.
FEATURED WORK
Evidence with its limits visible.
SAFETY RESEARCH · POSTER PREVIEW
Clinical-role claims and procedural access
A two-stage evaluation of text masking and unverified clinical-role framing. The public preview describes the question; detailed methods and results remain withheld in the source.
METHODS · OPEN-ACCESS PREPRINTS
Governance and reproducible reporting
Related RheumaView work on deterministic radiographic reporting and audit-ready provenance. Methodological background with a distinct scope.
A physician-founded research ecosystem.
RheumaView AI Safety is part of the RheumaView™ platform, connecting original research with independent evaluation and educational development. RheumaView presents the imaging platform; RheumaView Research hosts extended methods, publications, and demonstrations.
Rheumatology in Perspective
By Olga Goodman, MD · A RheumaView publication. The physician-authored publication at dr.rheumaview.com is another part of the RheumaView™ platform.
Put a clinical safety question to the test.
Define the system, the intended user, and the behavior you need to understand.
Professional research and education. This site does not provide patient-specific medical care. Scope & disclaimer