The United Nations convened its first-ever Global Dialogue on AI Governance in Geneva last week. The event was described as evidence-based. The evidence was presented. The industry that produced the evidence was also there.
The Number
One child per classroom. That is the figure. In eleven countries across the Global South, researchers found that one child per classroom had been targeted by AI-generated sexually explicit imagery of themselves.
The companies that built the tools used to produce this material were represented at the summit. Their position, broadly, was that innovation requires careful study.
Sonia Livingstone of the Independent International Scientific Panel on AI told the assembled delegates that the evidence for AI-linked harm was "much more compelling than the benefits." The assembled delegates thanked her for her contribution.
The Demographics of Harm
Ninety-nine percent of AI-generated deepfakes are sexual in nature. Ninety-six percent target women and girls. The remaining four percent are apparently something else entirely; the industry has not felt compelled to elaborate.
One in four women human rights defenders reported experiencing AI-assisted online violence in the past year. Women constitute thirty percent of the technology workforce. Eighty-eight percent of leading AI researchers are male.
These facts are presented here in sequence. The reader may draw their own connections. The industry has not.
The Question That Answered Itself
UN High Commissioner for Human Rights Volker Türk stood before the assembled representatives of the global technology sector and asked who, legally, is responsible when artificial intelligence causes harm.
The room contained the answer. Nobody said it.
Türk noted that safety standards exist for medicines, automobiles, and commercial aircraft — products whose manufacturers are required, by law, to not kill people at scale before the relevant committees have time to meet. He implied that AI might benefit from similar arrangements. The implication was received without verbal response.
2027
The summit will reconvene in New York in May 2027. This is nineteen months from now. In the intervening period, the tools that produced the evidence presented in Geneva will continue to be sold, subscribed to, updated, and optimized.
The children in those eleven classrooms will be a year older. The industry will have more data. The committees will have more to discuss.
---
*Sources: UNESCO — Global Dialogue on AI Governance · UNRIC — Protecting Children in the Age of AI*
