OpenAI says it signed the Charter of Engagement for Safer AI for Children and committed $2 million to a global research network on AI and children. The company announced it on November 12, 2026, in a post from its Global Affairs team. The signing took place earlier in the week at the Paris Peace Forum.
The amount is small next to the cost of a frontier training run. The signal is larger. The money buys measurement tools, and measurement is the missing piece in every child-safety rule written so far. This post explains what OpenAI said, how the coalition around the charter works, and what remains unknown.
TL;DR: what was announced #
| Question | Answer |
|---|---|
| Who announced it? | OpenAI, Global Affairs, November 12, 2026 |
| What did OpenAI sign? | Charter of Engagement for Safer AI for Children, at the Paris Peace Forum |
| How much? | $2 million to the research network under the Coalition for Children's Rights and Protection in the Age of AI |
| What does it fund? | Shared tools to measure AI risks and opportunities for children |
| Who controls findings? | Research teams keep their conclusions; results get published, including negative ones |
| Is it binding? | No. Voluntary commitments, plus OpenAI's support for future legislation |
What did OpenAI actually say? #
AI safety guardrails keeping a ball on a track, illustrating child AI safety research and measurement
The OpenAI post, titled "Supporting research on AI and children's safety," is short. It makes four claims.
First, OpenAI signed the Charter of Engagement for Safer AI for Children at the Paris Peace Forum earlier this week. Second, it committed $2 million to the scientific research network under the Coalition for Children's Rights and Protection in the Age of AI. Third, the contribution supports "shared measurement work," meaning tools researchers can use to measure the risks and opportunities AI poses for children. Fourth, the charter protects independence: research teams retain control over their conclusions, and findings will be published, including negative or inconclusive results.
The post also repeats OpenAI's policy line. It supports comprehensive youth AI safety legislation with age-appropriate safeguards by default, tools for parents, regular risk assessments and mitigation, and meaningful accountability. It says OpenAI looks forward to engaging with European policymakers as the EU Kids Act develops.
What the post does not say is just as useful. It names no research teams, gives no grant schedule, lists no other signatories and does not publish the charter text. Treat the $2 million as a pledge announced by the company, not as a funded program with a published budget.
What is the coalition behind the charter? #
The charter sits inside a government-led effort. According to the Élysée's page on the roadmap, the Coalition for Children's Rights and Protection in the Age of AI was launched by France, Spain and Kenya on July 7, 2026, at the first UN Global Dialogue on AI Governance. Australia joined later, and the four act as co-chairs.
The roadmap for 2026 and 2027 was advanced on September 22, 2026, during the UN General Assembly's high-level week. The page says the coalition intended to present its first voluntary commitments at the Paris Peace Forum on November 10 and 11. That is the event where OpenAI says it signed.
Membership numbers differ on the page itself. The body says 23 states, while the subtitle says 26 states and eight UN entities. The page does not explain the gap. The French outlet Banque des Territoires reports 23 states and eight UN entities, so treat 23 as the better-supported figure and 26 as unexplained.
What does the coalition want?
The stated goals, per the Élysée page:
- Ensure every child can benefit from AI under conditions that protect their safety, rights and development.
- Over time, subject AI systems that claim child benefits to pre-authorisation before market release.
- Build the evidence base and shared assessment methods that pre-authorisation would need.
- Meanwhile, seek faster voluntary action, including independent researchers' access to developers' models and data.
Work is split into thematic groups: safety and privacy by design, benchmarking and accountability, fighting AI-generated child sexual abuse material and deepfakes, and AI literacy. The page also describes a global network of research centres from the public, academic and private sectors, and invites companies and foundations to fund independent research and to open models for independent evaluation.
That last point links the pieces. The pre-authorisation idea only works if someone can measure what a model does to a child. OpenAI's $2 million is aimed at the measurement layer, which is the precondition.
Why measurement is the hard part #
Today, claims about AI and children rest on a mix of company policies, incident reports and lawsuits. There is no shared, accepted test that says a chatbot is safe for a 13-year-old. Without one, regulators can ban or allow, but they cannot certify.
Macron's statement on the Élysée page makes the point in political terms: "We cannot repeat the mistakes of the past and wait twenty years to see the impact of chatbots on minors." He adds that AI must not reach children "without first being tested, evaluated, and therefore authorized." A shared toolkit built before a harm becomes widespread is the practical version of that sentence.
Independence is the other constraint. A benchmark paid for by one lab invites doubt. The charter's reported terms, with teams owning conclusions and publishing negative results, are meant to answer that. They are OpenAI's description. Until the charter text and the network's governance are public, you cannot check how strong the protection is.
How does this fit OpenAI's earlier youth-safety moves? #
This is the latest step in a sequence. In a June 2, 2026 post, OpenAI called for an international youth AI safety institute ahead of the G7 Leaders' Summit in Évian. It laid out principles: privacy-preserving age estimation with protective defaults when age is unclear, annual youth safety risk assessments, parental controls over memory, data use and time limits, published safety policies, clear protocols for self-harm and exploitation, protection of minors' personal data, and independent audits under common standards.
The same post said OpenAI had strengthened safeguards for users under 18, launched parental controls with proactive notifications, and advanced age-prediction systems. It also pointed to a Model Spec section with dedicated principles for under-18 users.
The November pledge moves from principles to funded measurement. It is a different kind of commitment: money for evidence rather than a list of rules for others to follow.
The legal and reputational backdrop #
OpenAI is not acting in a calm environment. Our earlier coverage tracks the pressure:
- [Eight states sued OpenAI](https://explainx.ai/blog/eight-states-sue-openai-chatgpt-children-iowa-kansas-nebraska-2026) over ChatGPT and children.
- [Texas Attorney General Paxton sued OpenAI](https://explainx.ai/blog/texas-paxton-sues-openai-altman-chatgpt-minors-third-party-evaluator-2026) and raised the idea of a third-party evaluator.
- [Common Sense Media rated ChatGPT for teens](https://explainx.ai/blog/common-sense-media-chatgpt-for-teens-unacceptable-risk-crisis-referrals-2026) as an unacceptable risk, citing crisis referrals.
- [New York's governor signed a chatbot-minors law](https://explainx.ai/blog/hochul-signs-ai-chatbot-minors-law-baby-monitor-security-s9051b-2026) .
Read against that, the pledge is also a position in a fight over who sets the yardstick. A neutral, published measurement toolkit gives regulators something to point to, and gives the company a standard it helped shape. Both can be true. Judge it by what gets published.
Other firms are moving too. Meta rolled out global AI age estimation for Instagram teen accounts, and OpenAI's own teen usage report shows how young users actually use ChatGPT.
Parent and child shapes sharing one green glow, illustrating children and AI safety coalition work
What to watch next #
- The charter text. Whether independence and publication terms are written as binding commitments.
- Other signatories. Whether Google, Anthropic, Meta and others commit money or model access. OpenAI's post names none.
- The first deliverable. A benchmark, a protocol or a dataset that outside researchers can run. That is the proof the money produced something.
- EU KIDS Act progress. Banque des Territoires notes the European Commission proposed it on September 17, with a safety-by-design focus. OpenAI says it will engage as it develops.
- Access terms. The coalition asks companies to open models and data for independent evaluation. How much access any lab grants is the real test.
What this means for builders and parents #
If you build products that children may use, assume that measurement will arrive before rules do. Age estimation, protective defaults, parental controls and published safety policies are already the shared vocabulary in OpenAI's June principles and the coalition's goals. Design for them now.
If you are a parent or teacher, nothing changes today. The charter is voluntary, the research network is just forming, and no pre-market authorisation exists. Keep using the controls that products already offer, and watch for the first published results.
For OpenAI's other recent safety work, see its [community safety report on violence prevention and teen limits](https://explainx.ai/blog/openai-community-safety-report-violence-prevention-chatgpt-end-chats-teen-weapons-limits-2026); for the policy side, see how [Australia is moving on frontier AI rules](https://explainx.ai/blog/australia-mandatory-ai-incident-reporting-safety-regulator-november-2026), one of the coalition's four co-chair countries.
Bottom line #
OpenAI's announcement is real, small and specific: a signed charter, a $2 million pledge and a promise of independent, published research. It is not a rule, and it is not yet a result. The coalition's pre-authorisation goal depends on measurement tools that do not exist yet, which is exactly where the money is aimed. The next few months will show whether the charter produces tests anyone can run.
Related reading #
Details reflect sources available on November 12, 2026. We could not read the charter text itself, and figures may change as the coalition publishes more.