Final Stretch, High Stakes: Getting UN AI Governance Back on Track

Negotiations are advancing rapidly on the terms of reference (ToR) and modalities for two key initiatives established under the Global Digital Compact—the Independent International Scientific Panel on Artificial Intelligence and the Global Dialogue on Artificial Intelligence Governance. However, serious concerns have emerged about whether these mechanisms will live up to their promise of promoting a human rights-based, transparent, and effective approach to global AI governance.

Following initial opportunities for stakeholder input in February, the process moved through a series of revisions to the draft terms, beginning with a zero draft, followed by Rev1 and Rev2. While Rev1 introduced important improvements, such as the proposal for an Independent Appointment Committee to help oversee the selection of Scientific Panel members and a fully multistakeholder Global Dialogue, Rev2 marked a clear regression. The latest draft proposes a more state-centric appointment process and scales back stakeholder engagement to a single plenary segment, significantly weakening earlier commitments to transparency and participation.

The chapeau of the resolution now limits the mechanisms’ scope to non-military applications, risking the exclusion of many dual-use AI systems that carry clear human rights implications. Despite some strengthened language on human rights elsewhere in the document, the absence of clear connections to UN accountability mechanisms like OHCHR, the UPR or relevant Special Procedures further raises questions about the framework’s utility to protect human rights.

Since the outset, GPD has consistently advocated for these new initiatives to be anchored in the international human rights framework, and to include adequate resources for civil society engagement. Together with ECNL, we published detailed recommendations and have tracked the shifting contours of the negotiation text to assess whether those principles are being upheld.

While the formal consultation phase has concluded, ongoing negotiations on the final text remain difficult. G77 and China have stated they are not in a position to support the current draft, calling instead for institutional arrangements that prioritise support for developing countries in shaping national AI policies, legislation, and regulatory frameworks. Their position centers on the primacy of member states in global AI governance, offering limited space for meaningful stakeholder participation. At the same time, it emphasises the need for capacity building to ensure more equitable distribution of AI benefits.

The risks embedded in the current draft remain real—and the weeks ahead will be critical. Without strong institutional safeguards and meaningful follow-through, these mechanisms could fall short of their original mandate—or worse, contribute to legitimising a model of AI governance that lacks genuine participation or accountability. As the UN prepares to finalise the AI mechanisms modalities during next week’s last round of informal negotiations, we urge all stakeholders to call on member states to address the current gaps in the negotiated text. In particular, we recommend:

• Reinstating a broader role for stakeholders in selecting Scientific Panel members. A purely state-led selection process undermines inclusive, transparent and balanced governance.
• Mandating a Global Dialogue on AI that meaningfully involves all relevant stakeholders. This platform should support open, transparent and inclusive debate, avoiding dominance states or corporate interests. It should stay away from purely state negotiated outcomes.
• Strengthening links with existing UN human rights accountability mechanisms, including OHCHR, the UPR process, and relevant Special Procedures, to ensure alignment and accountability.

These adjustments are essential if the UN’s AI governance mechanisms are to deliver on their promise: to uphold human rights, foster meaningful inclusion, and shape effective, transparent global governance of AI for the benefit of all.