Q: What is the American Hospital Association asking the FDA to do about AI-enabled medical devices? A: The AHA is urging the FDA to develop measurement and evaluation frameworks that track AI tool performance across the product lifecycle and align with a risk-based approach to patient safety. These AI-enabled medical device evaluation standards would emphasize post-market monitoring, clear response triggers, and vendor responsibility to maintain tool integrity. Q: Why are new evaluation standards for AI tools considered urgent? A: Most AI tools today are evaluated on retrospective or static benchmarks that may not predict behavior in dynamic, real-world clinical settings, and changing patient populations and workflows can cause model drift. AI-enabled medical device evaluation standards would help detect bias, hallucination and drift early and reduce risk to patients. Q: What specific recommendations did the AHA make for these standards? A: The AHA recommended risk-based monitoring scaled to potential harm, alignment with the FDA’s total product lifecycle model, vendor accountability for ongoing integrity, and clear post-market evaluation plans with performance metrics and thresholds. It also suggested exploring streamlined 510(k) pathways when robust post-market monitoring is provided and noted some exclusions for clinical decision support and administrative tools. Q: How could these standards affect hospitals and frontline staff? A: Standards could give hospitals clearer rules for when to escalate, retrain or pause an AI tool and may shorten the time to access evidence-based, safe technologies if robust post-market plans are in place. They could also reduce rework and staffing strain by aligning monitoring with clinical workflows and by supporting training, technical assistance and potential grant funding for under-resourced systems. Q: What real-world risks do the proposed standards need to address? A: The standards need to address bias against underrepresented groups, hallucinated or incorrect outputs, and model drift that reduces accuracy over time. They should also cover human-AI interactions, explainability and requirements for human oversight so clinicians can spot errors and use tools safely. Q: What steps has the FDA already taken related to evaluating AI-enabled devices? A: The FDA published a request for public comment seeking feedback on practical approaches to measuring and evaluating AI-enabled medical devices in the real world, including metrics, post-market data sources and monitoring triggers. The agency also announced optional deployment of agentic AI capabilities for staff to support reviewers and investigators in complex workflows. Q: How would vendor responsibilities change under AHA’s proposed monitoring rules? A: Vendors would be expected to maintain the ongoing integrity of their products, provide defined performance metrics and thresholds, and communicate updates or incidents to users under post-development standards. Strong post-market evaluation and monitoring plans could also enable vendors to pursue streamlined 510(k) clearances under fewer applications when appropriate. Q: What practical steps can hospitals take now to prepare for these standards? A: Hospitals can inventory AI tools, document intended use and risk level, define metrics like accuracy and bias with preset drift triggers, and build monitoring plans that include regular audits and rapid response steps. They should keep a human-in-the-loop for clinical decisions, set vendor SLAs for performance and incident reporting, and train staff while collecting user feedback.