Artificial intelligence in health technology assessment: A commentary on promise, pitfalls, and governance

Generative and predictive Artificial Intelligence (AI) have moved from the periphery to the center of the methodological debate in Health Technology Assessment (HTA) and economic evaluation. Large language models now draft evidence summaries, machine-learning classifiers screen thousands of records in minutes, and emulators accelerate decision-analytic simulation—capabilities that arrived faster than the field's appraisal norms could adapt. This commentary reacts to that development and argues a deliberately dual thesis: AI can materially improve the efficiency, timeliness, and reach of HTA, but only if its adoption is disciplined by governance that treats validity, transparency, and equity as non-negotiable issues. Mapping AI methods onto the HTA workflow, I identify where value is most plausible—evidence identification and living reviews, evidence synthesis, real-world evidence and predictive modelling, decision-analytic modelling, and horizon scanning—and set against each the corresponding threats to validity, including data and label bias, opacity and irreproducibility, poor generalizability, unquantified model uncertainty, algorithmic unfairness, and automation-driven de-skilling. I contend that existing instruments (the National Institute for Health and Care Excellence evidence standards framework, ISPOR good-practice and generative-AI reports, reporting guidelines such as TRIPOD+AI and CHEERS 2022, and emerging regulatory principles) form a usable but incomplete scaffold, and that HTA bodies, researchers, and journals should converge on a governance-by-design model built on transparency, validation, human accountability, and standardized reporting. Particular attention is paid to low- and middle-income countries, where AI could widen or narrow the assessment capacity. I conclude with concrete recommendations for practice, research, and editorial policy.