AI Safety Needs a Human Rights Compass, Not Just Industry Rules

AI Safety Needs a Human Rights Compass, Not Just Industry Rules Leaders convene to examine AI governance alongside human rights. The discussion highlights responsible technology decision-making.

The debate over AI safety is often framed around technical control, model performance and future risk. Yet a central issue is closer to the present: how systems affect rights that are already protected under international standards. As companies race to define best practices for artificial intelligence, relying only on voluntary industry principles risks leaving key public interests outside the room.

The existing human rights framework offers a broader starting point. The United Nations system recognises rights linked to privacy, equality, freedom of expression, work, education and access to justice. These areas are directly touched by automated decision-making, data collection and generative technologies. The issue is not whether AI requires governance, but whether that governance should be built from scratch by the sector most affected by commercial incentives.

Human rights standards do not remove the need for technical expertise. They create a baseline for assessing harm, responsibility and remedy. The Office of the UN High Commissioner for Human Rights has long promoted principles requiring states to protect rights and companies to respect them, including through due diligence. Applied to AI, that approach shifts attention from abstract assurances to concrete questions: who may be harmed, what safeguards exist, and how people can challenge damaging outcomes.

This matters because the language of safety can be narrow. A system may be considered safe in an engineering sense while still producing discriminatory results, enabling intrusive surveillance or limiting access to essential services. A rights-based approach links technology regulation to accountability and public oversight rather than leaving it to private codes of conduct. It also places affected individuals and communities at the centre of evaluation, instead of treating them as downstream users of tools designed elsewhere.

The challenge for policymakers is to connect innovation with obligations that already exist. International human rights law was not written for large language models, but its principles were designed to endure across technological change. For the AI industry, adopting that framework would mean moving beyond self-defined safety benchmarks toward responsibilities that can be assessed by governments, courts, regulators and civil society. The result would be a more durable basis for trust in human rights and technology alike.