SafetyRisk, alignment & guardrails
OpenAI commits $2 million to children's AI safety research network after signing Paris-backed charter
OpenAI says it signed a voluntary charter on safer AI for children and will give $2 million to a government coalition's research network. The money is part of a roughly €8 million industry pool to build shared tools measuring AI's effects on young people.
From the OpenAI News article of November 12, 2026 and cited sources: OpenAI's $2 million is self-reported, and other outlets repeat it without independent confirmation. [1] [2] [6] [9]
SOURCES[1] OpenAI: Supporting research on AI and children's safety openai.com[2] Tech Times: AI Safety for Children pledge naming no tester… techtimes.com[6] Inside AI News: OpenAI pledges $2M insideai.news
AT Pulse research deskPublished · Updated · 5 min read · AI-assisted analysisAI-assisted, not human-edited. Methodology Corrections Who owns this site
- What happened
- OpenAI's $2 million is self-reported, and other outlets repeat it without independent confirmation. [1] [2] [6] [9]
- Original source
- OpenAI News article, company source
- Brief by
- AT Pulse research desk AI-assisted; publisher and contact on the desk page
- When
- November 12, 2026 article published
- Covered
- Nov 13 edition
- Evidence
- 10 sources across 10 sites
- 1OpenAI: Supporting research on AI and children's safetyopenai.com
- 2Tech Times: AI Safety for Children pledge naming no tester…techtimes.com
- 3France Diplomatie: Launch of the Coalitiondiplomatie.gouv.fr
- 4Élysée: Launch of the coalition roadmapelysee.fr
- 5Paris Peace Forum: Children First Declarationparispeaceforum.org
- 6Inside AI News: OpenAI pledges $2Minsideai.news
Reach: not measured yet
OpenAI's $2 million is self-reported, and other outlets repeat it without independent confirmation. It sits within a reported pool of about €8 million from AI developers including Anthropic. There is no per-company breakdown, no indication of whether the money is new, and no spending timeline. [1] [2] [6] [9]
The work focuses on measurement. Most evaluations assume adult users and score refusals and accuracy. Children need tests of age-appropriateness, emotional handling and privacy, such as age-grouped prompts, scoring rubrics, severity levels and rules for when a model should answer, redirect or refuse. A coalition working group covers benchmarking, but no method, dataset or benchmark has been published. [1] [4] [8] [6]
The pledge comes from a fast-growing policy effort. France, Spain and Kenya launched the coalition in July. It had 26 states by the UN General Assembly in September, where its roadmap said AI aimed at children should eventually need approval before reaching the market. The broader Children First Declaration has 77 signatories. [3] [4] [5] [10]
The commitments are voluntary. They name no testing standard, independent evaluator or enforcer. The French AI safety centre CeSIA says the text sets no common risk thresholds and does not guarantee testers the access they need. Some coverage also flags that the companies whose products would be tested are helping fund the research. [2] [6]
So far there is only a voluntary charter and modest funding, with no tools yet. If the benchmarking group produces child-specific test suites, they could underpin future pre-market approval rules.
Read the full assessment
Teams serving minors may want age-grouped testing and audit-ready risk assessments now.
Executive brief
OpenAI's $2 million is part of a pooled research pledge of about €8 million from AI developers, including Anthropic. The money comes alongside a voluntary charter. Analysts note that the charter does not yet say who will test AI products for children or who would enforce the rules (Tech Times). On 12 November 2026, OpenAI said it signed the Children First Declaration. It is funding a research network run by a coalition of governments to build shared tools that measure AI's risks and opportunities for young people (OpenAI). Only a feed excerpt of OpenAI's post could be retrieved. Details of the post come from secondary coverage.
What changed and event timeline
Coalition launched
France, Spain and Kenya launched the Coalition for Children's Rights and Protection in the Age of AI next to the first UN dialogue on AI governance. Its stated risks include emotional dependence, inappropriate content and effects on cognitive development ().
Roadmap at the UN General Assembly
The coalition grew to 26 states and 8 UN entities, with Australia joining as co-chair. It asked companies to fund independent research and open their models to evaluation.
More detail
It also said AI aimed at children should eventually need approval before reaching the market ().
Children First Declaration
The declaration was launched at the Paris Peace Forum's Child Summit. It has 77 signatories: 51 countries, 7 international organisations and 19 other partners. It covers children's welfare broadly, and the text names no funding amounts ().
€8M pledge reported
AI developers including OpenAI and Anthropic pledged about €8 million for research on AI and child development. Reporting did not say how the money is split, whether it is new funding, or when it will be spent ().
OpenAI's $2M
OpenAI said it signed the Charter of Engagement for Safer AI for Children and committed $2 million to the coalition's research network (;).
Capabilities and access
- No model, product or version is named. This is a research-funding and governance commitment, not a product release.
- The funded output is shared measurement tools for researchers (OpenAI).
- Secondary coverage says the charter covers age-appropriate safeguards on by default, parental controls and regular risk assessments (StartupHub.ai).
Read the full section
- No model, product or version is named. This is a research-funding and governance commitment, not a product release.
- The funded output is shared measurement tools for researchers (OpenAI).
- Secondary coverage says the charter covers age-appropriate safeguards on by default, parental controls and regular risk assessments (StartupHub.ai).
- There is no public timeline for when the tools will be available or how to access them.
Technical analysis for researchers and developers
- Evaluation gap: most AI evaluations assume adult users and score refusals and accuracy.
- Useful outputs would include test prompts grouped by age, scoring rubrics, severity levels, and rules for when a model should answer, redirect or refuse (Ken Ashe).
- Coalition structure: working groups cover safety, privacy, benchmarking, accountability and AI-generated child abuse material, plus a shared repository of research (Élysée).
Read the full section
- Evaluation gap: most AI evaluations assume adult users and score refusals and accuracy. Children need tests for whether answers suit their age, handle emotion well and protect privacy (Ken Ashe).
- Useful outputs would include test prompts grouped by age, scoring rubrics, severity levels, and rules for when a model should answer, redirect or refuse (Ken Ashe).
- Coalition structure: working groups cover safety, privacy, benchmarking, accountability and AI-generated child abuse material, plus a shared repository of research (Élysée).
- Reproducibility: research teams reportedly control their own conclusions and must publish negative or inconclusive results (Inside AI News). No method, dataset or benchmark has been published yet.
Claims and evidence
- $2M from OpenAI: this comes from OpenAI itself. Other outlets repeat it, and none independently confirms the amount (OpenAI; TokenPost).
- Research independence and publishing negative results: these are charter terms as reported. No independent party has checked how they are enforced (StartupHub.ai).
- €8M industry pool: reported by Tech Times, with no breakdown by company (Tech Times).
Read the full section
- $2M from OpenAI: this comes from OpenAI itself. Other outlets repeat it, and none independently confirms the amount (OpenAI; TokenPost).
- Research independence and publishing negative results: these are charter terms as reported. No independent party has checked how they are enforced (StartupHub.ai).
- €8M industry pool: reported by Tech Times, with no breakdown by company (Tech Times).
- Signatory counts: confirmed by the Paris Peace Forum (Paris Peace Forum).
Context and prior work
- The coalition grew out of France's G7 focus on protecting children online.
- French press describe the goal as a form of prior approval for children's AI tools before they go on sale (Banque des Territoires).
- The EU is still negotiating a Kids Act alongside this effort (StartupHub.ai).
Read the full section
- The coalition grew out of France's G7 focus on protecting children online. It also lines up with the UN Secretary-General's call for an "AI Child Safety Pledge" (France Diplomatie).
- French press describe the goal as a form of prior approval for children's AI tools before they go on sale (Banque des Territoires).
- The EU is still negotiating a Kids Act alongside this effort (StartupHub.ai).
Limitations, safety and contested findings
- No enforcement: the commitments are voluntary. The text names no testing standard, independent evaluator or approval authority.
- Criticism from CeSIA (French AI safety centre): it says the text sets no common risk thresholds and does not guarantee independent testers the access they need (same source).
- Conflict of interest: the companies whose products would be tested are helping to fund the research (Tech Times).
Read the full section
- No enforcement: the commitments are voluntary. The text names no testing standard, independent evaluator or approval authority. The shared testing framework does not exist yet (Tech Times).
- Criticism from CeSIA (French AI safety centre): it says the text sets no common risk thresholds and does not guarantee independent testers the access they need (same source).
- Conflict of interest: the companies whose products would be tested are helping to fund the research (Tech Times). Industry-funded research has a mixed record on independence (Inside AI News).
- Scale: $2M is small next to public funding for child-development research (Inside AI News).
- Membership counts conflict: sources give 18, 23 and 26 states at different dates.
Business and practitioner implications
- Watch the benchmarking working group. If it produces child-specific test suites, they could become the basis for any future approval-before-launch rules (Élysée).
- Teams with products used by minors can test against age groups and keep audit-ready risk assessments.
- The value depends on whether the tools change products before launch, not on the size of the grant (Ken Ashe).
Read the full section
- Watch the benchmarking working group. If it produces child-specific test suites, they could become the basis for any future approval-before-launch rules (Élysée).
- Get ready now. Teams with products used by minors can test against age groups and keep audit-ready risk assessments. Charter partners reportedly include Anthropic, Google and Microsoft (Tech Times).
- Tools matter more than the money. The value depends on whether the tools change products before launch, not on the size of the grant (Ken Ashe).