Lilith.
⌕
Editorial illustration: OpenAI puts $2 million behind shared measures of AI’s impact on children
Lilith illustration · editorial remix

At the Paris Peace Forum, OpenAI signed the Charter of Engagement for Safer AI for Children and pledged $2 million to the scientific network under the Coalition for Children’s Rights and Protection in the Age of AI. The money is intended to support shared tools that let researchers measure AI’s risks and opportunities for young people on comparable terms.

The $2 million funds a common yardstick, not another parental toggle

The announcement is not a new ChatGPT feature. OpenAI is funding measurement work meant to inform safer products and policy later. The company says the charter establishes principles for safe, age appropriate AI, a research programme and safeguards for independence.

Research teams are supposed to retain control over their conclusions. Negative and inconclusive findings are to be published as well. OpenAI also links the initiative to the developing EU Kids Act and advocates safeguards by default, parental tools, regular risk assessments and meaningful accountability.

A shared yardstick could replace arguments fought through press releases

Developers can add parental controls and safety filters today without using the same method to evaluate outcomes. Shared tools could make it possible to compare whether safeguards reduce specific harms while preserving useful access to learning and creative tools.

Regulators need the same discipline. Repeatable measurements provide a stronger basis for rules than isolated incidents or metrics selected by each vendor. The programme’s value therefore lies less in the cheque than in whether it produces a public method others can use.

Vendor funding will test the research network’s promised independence

OpenAI has not yet named recipients, a timetable or the process for choosing what will be measured. The charter promises independent conclusions, but funding from a company with its own youth facing product creates an obvious conflict that transparent governance and published methods must address.

Publishing negative or inconclusive findings is a useful commitment. Without open protocols, sample descriptions and reproducible measurements, outsiders will not know whether this is a shared standard or another set of internal indicators.

Open methods and inconvenient findings will reveal who the network serves

The first signals will be the selected research teams, their selection rules and public methodologies. The second will be findings that are uncomfortable for participating companies yet appear without being softened.

The investment will matter if independent academics, rival companies and European policymakers can use the tools. Only then can $2 million buy a common language for safety rather than merely a seat for OpenAI at the table.

Lilith's verdict

Researchers have been handed a ruler by one of the companies they must measure. Credibility begins when that ruler produces a number OpenAI would rather not put on its homepage.