Health information professionals are stepping into a critical role as gatekeepers against AI-generated medical errors, armed with a new ethical framework designed specifically for clinical environments where algorithmic mistakes can have life-or-death consequences. The Medical Library Association (MLA) has released BEST-P, a comprehensive guide addressing bias, expertise, sustainability, transparency, privacy, and property concerns in generative AI use within healthcare. Unlike broader AI ethics discussions, this framework tackles a specific vulnerability in medicine: AI systems that generate plausible-sounding but dangerously inaccurate clinical advice.

Why Is AI Accuracy Such a Critical Problem in Healthcare?

Generative AI models are designed to produce fluent, confident-sounding responses, but they prioritize language smoothness over factual precision. In healthcare settings, this creates an acute danger. When AI-generated literature summaries or clinical decision support outputs contain subtle errors, they can lead to misdiagnosis, inappropriate treatment, or flawed health policy decisions that affect patient safety. The stakes are fundamentally different from AI errors in other sectors; a recommendation algorithm that suggests the wrong movie is inconvenient, but an AI system that confidently provides incorrect clinical guidance can be fatal.

The MLA framework emphasizes that health information professionals must serve as essential human oversight layers, using their domain expertise to critically evaluate and verify AI-generated evidence against trusted sources. This is not about rejecting AI tools outright, but rather about establishing rigorous verification workflows before any AI-generated content enters clinical decision-making processes.

What Are the Core Ethical Challenges AI Poses in Medical Settings?

The BEST-P framework identifies six interconnected challenges that health information professionals must address:

• Bias Amplification: Generative AI models may amplify existing biases present in scholarly data and publications, translating skewed research into discriminatory clinical advice that directly risks misdiagnosis and inequitable care for marginalized populations.
• Accuracy and Expertise Gaps: AI outputs may contain incorrect, incomplete, or misleading information because the technology generates plausible-sounding responses without guaranteeing factual accuracy, requiring human domain expertise to verify claims.
• Environmental and Labor Costs: Generative AI requires significant natural resources including water and electricity for data centers, rare earth metals for hardware, and human labor, often from the global south, creating sustainability and equity concerns.
• Reproducibility and Transparency: Reproducibility is foundational to evidence synthesis and scientific method, requiring transparency and traceability of AI outputs so researchers can understand how conclusions were generated.
• Privacy and Data Protection: Generative AI models often retain user inputs for training purposes, creating risks that clinicians might inadvertently input re-identifiable patient data, violating privacy protections and patient trust.
• Intellectual Property Rights: Questions remain about how proprietary research and clinical knowledge are used in training AI systems, and whether proper attribution and consent are obtained.

How Can Health Information Professionals Mitigate AI Risks in Clinical Practice?

Rather than prescribing rigid rules, the MLA framework emphasizes competencies and skills that health information professionals already possess and can adapt for responsible AI use. These include information retrieval expertise, critical appraisal abilities, and evidence evaluation training. Health information professionals can employ several practical strategies:

• Bias Detection Training: Receive training and access to tools that help recognize bias in AI-generated content, then develop workflows to mitigate identified biases before clinical implementation.
• Prompt Engineering and Verification: Apply information retrieval skills to craft precise prompts that elicit more accurate AI responses, then verify outputs against trusted databases and peer-reviewed sources before clinical use.
• Transparency and Disclosure: Disclose when and how AI was used in evidence synthesis, confirm that humans remain ultimately responsible for all AI-generated content, and maintain clear documentation of verification processes.
• Sustainability Assessment: Evaluate whether the environmental and labor costs of using generative AI justify anticipated benefits, and advocate for vendor transparency regarding actual environmental impacts.
• Privacy Safeguards: Establish protocols to prevent re-identifiable patient data from being input into AI systems, and ensure that sensitive health information remains protected throughout the AI-assisted workflow.

The framework positions health information professionals not as obstacles to AI adoption, but as essential partners in responsible implementation. Their existing competencies in evidence-based practice, information stewardship, and ethical knowledge management directly address the gaps that pure algorithmic systems cannot fill.

What Does Responsible AI Governance Look Like in Practice?

The MLA framework recognizes that responsible AI governance requires institutional commitment beyond individual professional practice. Health information professionals should advocate for organizational policies that mandate human oversight of AI-generated clinical content, require disclosure of when AI was used in evidence synthesis, and establish clear accountability chains. This means institutions must invest in training staff to recognize AI-generated bias, provide access to verification tools, and create workflows that treat AI outputs as preliminary findings requiring expert validation rather than final recommendations.

The framework also emphasizes that health information professionals should recommend using generative AI only in situations where existing technologies or human expertise would not perform at the same level. This prevents unnecessary reliance on AI systems and ensures that the technology serves a genuine clinical need rather than simply automating processes for efficiency's sake.

By establishing these competencies and practices now, health information professionals can help ensure that generative AI becomes a tool that enhances evidence-based medicine rather than a source of hidden errors that undermine patient safety and clinical equity.

" }