Proposal
This policy proposal establishes labour, workplace safety, and social responsibility requirements for AIMSS-certified organisations. AI systems are often built and maintained by a broad network of employees, contractors, data annotators, content moderators, evaluators, red-teamers, outsourced vendors, and infrastructure providers. These workers may be exposed to harmful content, insecure working conditions, unfair compensation, excessive monitoring, or unclear accountability. Responsible AI certification must therefore include the people and communities affected by AI production, not only the technical system.
The organisation shall maintain an AI Labour and Social Responsibility Policy covering all workers who materially contribute to AI development, testing, deployment, monitoring, moderation, or evaluation. This includes direct employees, contractors, outsourced workers, and material supplier personnel where the organisation has influence through contract or procurement. The policy shall define minimum labour expectations, worker safety controls, harmful-content protections, grievance channels, supplier due diligence requirements, and escalation procedures.
For AI work involving exposure to harmful, violent, sexual, abusive, discriminatory, traumatic, or otherwise distressing material, the organisation shall implement specific psychological safety controls. These controls shall include informed role expectations, content exposure limits, access to mental health support, rotation or opt-out mechanisms where feasible, supervisor training, and incident reporting. Workers must be informed of the nature of the work before assignment and must not be penalised for raising safety concerns.
The policy shall require fair and transparent compensation practices. Organisations shall document how workers involved in data labelling, evaluation, testing, and moderation are compensated. Where work is outsourced, supplier contracts shall prohibit exploitative labour practices, unauthorised subcontracting, retaliation, and unsafe working conditions. Material suppliers shall be reviewed before onboarding and periodically thereafter. Supplier due diligence shall consider compensation, working hours, grievance channels, psychological safety, data security, and subcontracting practices.
The social responsibility component shall require organisations to assess how AI systems may affect different communities, cultures, languages, accessibility needs, and vulnerable groups. For high-risk or public-facing systems, the organisation shall evaluate whether outputs, decisions, or user experiences may reinforce harmful stereotypes, exclude disabled users, marginalise linguistic groups, or produce culturally inappropriate results. Where material risk exists, affected communities or representative stakeholders should be consulted in a structured and documented way.
The organisation shall maintain grievance and whistleblower channels for workers and affected stakeholders. These channels must allow confidential reporting of unsafe practices, unethical instructions, discriminatory outcomes, supplier abuse, or social harm. Reports shall be logged, investigated, and resolved according to defined timeframes. Retaliation against workers or stakeholders who raise concerns shall be treated as a serious certification issue.
Certification evidence shall include labour policies, supplier contracts, due diligence records, worker safety procedures, training records, grievance logs, harmful-content safeguards, compensation documentation, stakeholder engagement records, accessibility testing, cultural bias evaluations, and corrective action plans. Failure to conduct supplier due diligence or protect workers from foreseeable harm should be treated as a major nonconformity. Retaliation, forced concealment of harm, exploitative outsourcing, or repeated failure to address unsafe working conditions should be treated as a critical nonconformity.