Skip to content

Non-maleficence

Purpose: To ensure that AI tools are designed, deployed and operated in ways that actively reduce harm, minimise foreseeable risks, and avoid causing physical, psychological or institutional damage.

Organisational / Technical Measure
A. Harm identification and risk awareness
BOTH Potential harms associated with the AI system (direct and indirect) have been identified and documented before deployment of tool
ORG Harm assessments consider multiple dimensions of impact (e.g. safety, dignity, fairness etc)
BOTH Risks are reviewed not only at design stage but also when contexts, data or usage patterns change
ORG Organisation has a clear, accessible definition of “harm” in the system specific use context
B. Human responsibility and intervention
BOTH Human staff are aware of duties to intervene when the tool produces outputs that may cause harm
ORG Clear guidance exists on when and how human intervention should occur in response to harmful or unsafe outcomes
ORG Responsibility for monitoring harm indicators and responding to them is explicitly assigned
C. Monitoring, learning and continuous improvement
TECH Ongoing monitoring track indicators related to harm and unintended consequences over an extended time period
BOTH Feedback from affected users or stakeholders is systematically collected and reviewed
ORG Identified harms, near-misses or emerging risks lead to documented corrective actions and system improvements
ORG Risk mitigation practices are periodically reassessed

Source: AIOLIA deliverable 3.1