OECD Maps Rules for Artificial Intelligence in the Workplace

The OECD review shows that governments introducing artificial intelligence into the workplace are generally not creating entirely new labour law. They rely on existing rules covering privacy, non-discrimination, occupational safety and information for employees. Special measures mainly concern training, support for technology adoption and some forms of algorithmic management. This reduces the number of new laws, but increases the need to interpret several legal regimes together. The study covers G7 countries, the European Union and selected Latin American states. All G7 members already recommend or require measures against bias and discrimination. Public employment services in six countries use AI to match applicants with jobs or tailor support. The benefits are clear when the system broadens access to services. An error, however, may systematically push a person away from suitable employment. Human oversight remains a central requirement. The European Artificial Intelligence Act requires it for high-risk systems, while the Platform Work Directive limits automated decisions to suspend or terminate a worker's account. The OECD notes that it is not enough to place a person formally in the approval chain. That person must understand the evidence, have the authority to change the decision and bear clear responsibility for intervening. For employers, the highest-risk uses are recruitment, shift scheduling, performance evaluation, pay and dismissal. Responsibility cannot be transferred entirely to the software supplier. A company must know the system's purpose, the origin of its data, the indicators it monitors, the groups that may be disadvantaged and the complaints procedure. It also needs records of when a person changed a recommendation and whether errors recur for the same group. The OECD emphasises social dialogue and collective bargaining. Employees are often the first to recognise that an indicator measures the wrong activity or creates unsafe pressure. Involving trade unions or worker representatives therefore need not obstruct a project; it may reveal a problem before full deployment. The authors also warn that authorities must verify “whether measures are actually implemented and how effective they are”. A formal principle without examination of outcomes is insufficient. Before buying a system, a company should list the decisions it will influence and identify the possible harm in each case. It should then test the system on representative data, define human intervention, train managers and provide an accessible route to remedy. For mental and physical safety, indirect effects also matter: faster work pace, unpredictable shifts or continuous monitoring. The project's return should be calculated after these risks are deducted, not only from hours saved.

OECD Maps Rules for Artificial Intelligence in the Workplace

The full article is available to users registered at Hard Skills