Four AI Companies Agree to Outside Child-Safety Tests of Their Tools
OpenAI, Anthropic, Microsoft and Black Forest Labs have backed a Paris charter that lets independent researchers test how their AI affects children. It is voluntary, and research funding totals €8 million.
Four AI companies have agreed to let independent researchers test how their chatbots and image tools affect children. OpenAI, Anthropic, Microsoft and Black Forest Labs backed a new charter at the Paris Peace Forum this week. The organisers say backers will put a total of €8 million into research on AI and child development.
The charter was published on 10 November. The coalition behind it is led by France, Spain and Kenya, and ten United Nations agencies take part. It is a voluntary commitment, not a law, and it creates no legal duties.
What the companies promise
The text sets out ten principles. They include safeguards against overuse and emotional dependence, tools for parents, strict limits on the data collected about children, and age checks that do not rely only on a child typing in a date of birth.
Outside experts will lead the research. Companies may give them access to their systems, but paying for the work gives a company no right to block or delay the results. Findings will be published even when they are negative or inconclusive. Testers are told to check whether users under 18 are correctly identified, and to try realistic misuse, such as teenagers using indirect language or trying to get around safety settings.
OpenAI is giving $2 million of the total. The company said on Thursday that the money will fund shared methods for measuring the risks and benefits of AI for children.
Emmanuel Macron, the French president, asked for more than the charter offers. He said that no AI should reach children before it has been tested, evaluated and approved. His comments appeared in a statement from the summit organisers. The charter itself sets up no approval process. Google, Meta and xAI are not among the four companies that endorsed it.
Teenagers already use these tools
Young people are not waiting for the evidence. A paper published on Wednesday in a journal on adolescent health brings recent figures together. In one US survey of 1,060 teenagers, 72% had tried a chatbot built to act as a friend or companion. In a separate national study, 15% of teenagers who used such chatbots said one had encouraged self-harm. Only 11% of teachers had received guidance on what to do when a student’s use of AI seemed harmful.
Last month we reported that most teenagers in a large student survey turn to AI when they feel low. New York has already passed a law that makes chatbots check ages before they act like a friend.
Why it matters: Parents and schools have little independent evidence about how chatbots affect children. This charter could produce some, but it is voluntary and does not yet cover several of the largest chatbot makers.
