Governance, Deployment & Responsible Use of LMMs

WHO frames LMM governance across three stages of the AI value chain:

1. Development — developers train foundation models (data quality, bias, privacy, environmental impact)
2. Provision — providers fine-tune or integrate models into health applications
3. Deployment — hospitals, ministries, clinicians, or individuals use approved tools in practice

Responsibilities differ at each stage, but human rights and ethical principles are non-negotiable throughout.

Developer and provider obligations

Developers should conduct data protection impact assessments, engage diverse stakeholders (including clinicians and patients via "human oversight colleges"), design for accuracy, and label AI-generated content. Governments may require pre-certification, audits, target product profiles for health LMMs, and transparency on training data.

Providers adapting foundation models for clinical use share regulatory burdens — especially when changes diverge from the developer's original design. Impact assessments audited by independent third parties should address ethics, human rights, safety, and data protection across the lifecycle.

Deployer responsibilities (where you may sit)

As a deployer — a hospital, clinic, or individual clinician using an LMM product — WHO expects you to:

• Avoid inappropriate settings when biases or contextual limitations are known
• Communicate risks clearly — not buried in fine print; suspend use if harm occurs
• Improve accessibility — languages, pricing, and inclusion for underserved groups
• Train the workforce on decision-making, bias avoidance, patient engagement, and cybersecurity
• Use procurement power to demand transparency from vendors

Even after regulatory approval, LMM outputs may change unpredictably. Mandatory post-release audits and impact assessments are recommended when deployment scales.

Liability and international governance

As LMM-related harm occurs, governments may introduce presumptions of causality, strict liability, or no-fault compensation funds — balancing patient redress against discouraging beneficial innovation.

International cooperation is needed so governance keeps pace with global technology — shaped by all countries, not only high-income nations and large tech firms.

For Hong Kong healthcare professionals

Participate in institutional AI governance: document intended use, monitor outputs, report incidents, and maintain skills to practise safely if AI tools are unavailable. Your clinical judgement remains the final safeguard.

Source: WHO — Ethics and governance of artificial intelligence for health: Guidance on large multi-modal models (2024)