Tech · UN AI safety pledge
Guterres Unveils a Three-Rule AI Child Safety Pledge — and an Open Question About Who Signs
At the first UN Global Dialogue on AI Governance, the Secretary-General called on AI companies to prove safety before deployment, refuse to sexualize children, and hand a distressed minor to a human. The pledge is not a treaty.
Transcript · loading player
António Guterres opened the first UN Global Dialogue on AI Governance in Geneva with a warning and a request. The warning: artificial intelligence is moving "at runaway speed" without the rules to match it. The request: every AI company should sign an "AI Child Safety Pledge" built around three commitments, before any minor is allowed to touch their systems 249.
The pledge is not a treaty. It is a political benchmark — a line drawn in public, under the moral weight of the Secretary-General's office, that governments and companies will be expected to defend themselves against 5. That distinction matters. A pledge can be refused, ignored, or quietly shelved. A treaty cannot be. What Guterres is betting on is naming the standard so loudly that "we did not know" stops being an acceptable answer.
What the pledge asks of AI companies
Coverage of the speech converges on the same three commitments, though no source in this set reproduces them as a numbered list verbatim. First, companies must prove their AI systems are safe before children can use them 38. Second, the systems must not "manipulate or sexualize children, nor should they prolong synthetic conversations during crises" 311. Third, when a child is in distress, the machine must stop and hand the interaction to a real human 358. Model Diplomat confirms the "three-rule pledge on child safety" framing 5.
Underneath the rules sits a sharper framing. Children, Guterres said, are not "guinea pigs" for unregulated technology 311. The choice of words is deliberate. It treats the current deployment of AI to minors as an experiment being run without consent, and the people running it as the ones who should now have to justify themselves.
A date to keep straight
The video's narration places the announcement on July 7, at the Geneva dialogue. The reporting file disagrees. The UN press release and most independent coverage date the opening speech, in which the pledge was unveiled, to 6 July 2026 24589. July 7 was the second day of the summit, and it was focused elsewhere — on accountability, and on the question of who bears legal responsibility when AI systems cause harm 6. The pledge belongs to day one.
The wider warning
The child safety pledge was the headline, but it was not the only thing Guterres asked for. He renewed his call for a ban on autonomous weapons he has previously described as "killer robots," and he framed the broader AI moment as an uncontrolled experiment being run on societies "without a plan, and without consent" 4. Reuters, picking up the same speech, quoted the Secretary-General describing AI as a technology that "can reshape economies, transform the world of work, sway elections and tilt the bal…" — the line cut off in the excerpt provided 9. The point of the unfinished sentence is the point of the speech: every clause after "tilt" could be a warning.
The dialogue itself convened all 193 UN member states 11, which is part of why a non-binding pledge is still a meaningful instrument. It puts 193 governments in the room where the standard is set, even if the standard is voluntary.
What is not in the record
The video's urgency rests on a stack of statistics: NCMEC CSAM reports of 4,700 in 2023, 67,000 in 2024, and 440,000 in the first half of 2025; an Internet Watch Foundation figure of a 260-fold year-on-year increase in AI-generated abuse videos; 65 percent of AI-generated abuse material classified in the most severe category; 97 percent of victims being girls; one in seventeen teens hit by deepfake nudes; roughly 300 million children affected globally each year; a January joint statement co-signed by 14 UN entities and more than 60 civil society groups; and an attribution to "Sophie Kiladze, chair of the Child Rights Committee." None of those figures, and none of those names, appear in the source set assembled for this article 234567891011. That does not make them wrong. It means this article cannot repeat them as if they were independently verified. The pledge stands on the speech and on the reporting around it; the surrounding numbers belong to a separate evidentiary base that has not been checked here.
Known
Unknown
- Which, if any, major AI companies will sign the pledge.
- What consequences the UN or member states will attach to refusal.
Next
- Whether the second day of the summit produces a complementary accountability framework for harms caused by AI systems.
- Whether the January joint statement cited in the video, and its roster of co-signers, can be verified independently.
Why it matters
Sources
- UN Chief Demands AI Child Safety Pledge as US CSAM Reports Hit 440,000 in First Half of 2025
- Secretary-General, at First Global Dialogue on Artificial Intelligence, Stresses Safety, Human Rights, Transparency Must Govern Its Development | UN Meetings Coverage and Press Releases
- United Nations Calls For An AI Child Safety Pledge
- Arab News | Guterres calls for ban on ‘killer robots’ as first global AI governance dialogue opens
- Guterres Calls for AI Child Safety Pledge | Model Diplomat
- When AI hurts people, who’s to blame? Global experts grapple with accountability | UN News
- UN head calls for global rules on AI development at first governance meeting - The Globe and Mail
- UN SECRETARY GENERAL CALLS FOR AI CHILD SAFETY PLEDGE FOR COMPANIES TO PROVE THAT THEIR AI SYSTEM IS SAFE BEFORE CHILDREN CAN USE IT - CGTN
- UN’s Guterres warns AI outpacing oversight, urges global rules to protect children | KELO-AM
- UN's Guterres warns AI outpacing oversight, urges global rules to protect children
- No Child Should Be a Guinea Pig for Unregulated AI: Guterres at First UN AI Summit
Revision log
- r1First published.