UNESCO on Gender Equality in AI: Who Shapes the Rules?

The Brief

UNESCO warns that AI can reproduce and amplify gender bias.
Its brochure brings together research, policy work and programmes; it is not a new global survey.
Earlier UNESCO model tests found gender stereotypes in generated content. Those results concern the models tested at the time.
Employers should examine how AI affects hiring, training and decisions about people.
Women need opportunities to develop and govern AI, alongside routes to challenge harmful outcomes.

UNESCO’s commitment to gender equality in AI

There is something persuasive about a score on a screen. A person’s judgement invites questions; a software-generated ranking can look like the answer. UNESCO’s case for gender equality in AI asks us to examine that confidence. When a machine helps decide who looks capable, who gets considered and whose experience counts, we should want to know what it has learnt about women.

In its brochure, UNESCO’s Commitment to Gender Equality in AI, the organisation warns that AI can carry existing gender inequalities into its systems and outputs. It also presents women as innovators and decision-makers whose contribution is essential to developing better technology. That combination deserves attention: women have a stake in both the decisions AI makes possible and the authority to shape them.

For a workplace buying its next AI tool, this raises a question that rarely fits neatly into a product demonstration: what does the software treat as evidence of potential?

The past can look remarkably efficient

Consider a hypothetical employer using previous hiring records to help identify promising candidates. If those records reflect years of restricted access to certain jobs, learning their patterns could reproduce the restriction. A system might become very good at finding people who resemble earlier hires without becoming good at recognising overlooked talent.

For an Indian employer, useful testing questions might include how a tool handles career breaks, regional-language experience or qualifications from less familiar institutions. These are examples of what to investigate, not findings that every AI hiring system discriminates in these ways. The employer should establish whether its criteria measure the work someone can do.

Even apparently harmless choices deserve scrutiny. Our examination of the feminine voice used for digital assistants explores the associations built into the way technology presents itself. A product’s default can carry an assumption long before anyone makes a consequential decision with it.

UNESCO’s 2024 research offers a concrete illustration. It examined GPT-2, GPT-3.5 and Llama 2, finding gender bias in their generated content. In Llama 2’s output, women appeared in domestic roles four times as often as men. These were findings about particular models and tests, not a measurement of every AI product available in 2026.

Still, they give organisations a reason to check what they are accepting. Imagine an AI-assisted training exercise that repeatedly makes the manager a man and the assistant a woman. Each example may seem trivial. Together, they would narrow the picture of who belongs in charge. An editor or trainer can change that picture before it reaches an audience.

Who gets the time to learn?

The workplace implications extend beyond biased outputs. In March 2026, the International Labour Organisation reported that, across countries with available data, around 29% of female-dominated occupations were exposed to generative AI, compared with 16% of male-dominated occupations. It linked the imbalance to women’s concentration in clerical, administrative and business-support work.

Those figures describe occupational exposure. They do not mean that 29% of working women will lose their jobs, nor should they be presented as an India-specific estimate. How organisations redesign work remains consequential.

An employer introducing AI into administrative work should involve the people doing that work in deciding what changes. Which tasks can be reduced? Which responsibilities require judgement? What training will employees receive before expectations change?

Making that training available during paid hours is one practical choice. A course offered only after work asks employees to supply time that may already be committed elsewhere. Access to an account also tells us little about access to useful projects, mentoring or promotion. The barriers surrounding women’s participation in AI deserve attention throughout a career.

Women should have opportunities to specify requirements, evaluate products and lead implementation. Their participation should carry decision-making authority, with the organisation retaining responsibility for fair outcomes.

Put the questions in the purchase order

UNESCO’s brochure connects education and mentorship with research, ethical standards, safety and women’s leadership. Its wider Recommendation on the Ethics of Artificial Intelligence, adopted in 2021, provides a framework for governments addressing these questions. It should not be mistaken for a new workplace law.

We believe that employers can begin with the next purchase.

Before approving a tool that influences recruitment or progression, ask the supplier what it tested, which groups were represented, where performance differed and what remains unknown. A claim of fairness needs evidence relevant to the intended use.

Then test the actual workflow. Human review needs someone with the time, information and authority to overturn a recommendation. Merely placing a manager at the end of the process offers little reassurance if the score has already settled the discussion.

Applicants and employees also need a usable route to query consequential decisions. Organisations should explain where AI contributes, identify who handles concerns and check whether remedies work. Where lawful and appropriate, carefully protected, gender-disaggregated outcome data can help identify disparities requiring investigation. Personal information should be collected for a defined purpose, with safeguards and limits.

Finally, establish consistent expectations for employees using AI themselves. Our coverage of the gender penalty attached to AI-assisted CVs considers how people’s judgements can complicate adoption. A workplace needs clear standards for acceptable assistance and accountability, applied consistently to everyone.

UNESCO on gender equality in AI: The final word

At Change in Content, we see an opportunity in the moment before a workplace accepts an AI recommendation as routine. Someone still chooses the criteria, signs the contract and decides whether an objection deserves attention.

Those choices can widen opportunity. They can give an overlooked applicant a fair assessment, an employee time to learn and women the authority to shape tools that affect their careers. Progress should be judged by whether those opportunities become real. A faster decision is useful only when the organisation is prepared to stand behind it.

Editorial note and sources

This commentary draws on UNESCO’s brochure, its earlier research and ethics framework, and ILO reporting. Workplace scenarios are illustrative; no interviews or independent model tests were conducted for this article. Historical model results and occupational exposure estimates are identified in context. The practical recommendations are Change in Content’s editorial analysis, rather than legal advice or claims about a named employer.

Sources

UNESCO on Gender Equality in AI: Who Taught the Machine What a Woman Can Do?

UNESCO’s brochure connects AI with the inequalities around it. For workplaces, the urgent question is whose assumptions become automated decisions, and who gets to challenge them.