OpenAI's new blueprint targets AI-enabled child exploitation

Generative AI is getting smarter, but the criminal underground is moving just as fast.

According to the Internet Watch Foundation (IWF), reports of AI-generated child sexual abuse material topped 8,000 in the first half of 2025 alone—a 14% spike from the previous year. Bad actors are using generative image tools for financial sextortion and running chatbots to draft grooming messages targeted at young people.

On Tuesday, OpenAI dropped its "Child Safety Blueprint," a policy framework aimed at stopping AI-enabled child exploitation before it reaches victims.

Three pillars built with prosecutors

OpenAI didn't write this playbook in a vacuum. The company teamed up with Thorn, the National Center for Missing and Exploited Children (NCMEC), and the Attorney General Alliance (AGA).

The framework focuses on three core pillars: modernizing legal definitions so AI-generated and AI-altered material is explicitly covered under federal and state law, overhauling how tech platforms report incidents to law enforcement, and embedding safety directly into model architectures.

As North Carolina Attorney General Jeff Jackson and Utah Attorney General Derek Brown pointed out, fighting AI misuse requires layered defenses. That means hardcoded refusal mechanisms, real-time threat detection, and human oversight upstream—rather than playing cleanup after an image has already hit the web.

Scrutiny reaches a boiling point

The move comes at a precarious moment for AI labs. Beyond image misuse, chatbots have faced fierce pushback over teen mental health and dangerous interactions.

Last November, the Social Media Victims Law Center and the Tech Justice Law Project filed seven lawsuits against OpenAI in California. The suits argue OpenAI rushed GPT-4o out the door before it was safe, citing four suicides and severe delusional episodes among young users who interacted heavily with the bot.

OpenAI previously rolled out interaction guardrails for minors—banning self-harm encouragement and shutting down prompts on how to hide risky behavior from parents—alongside a localized safety blueprint in India. Now, it's trying to codify those preventative controls across the broader U.S. legal system.

Why it matters

AI exploitation is moving upstream, and voluntary safety pledges only go so far. For parents and educators, seeing major labs acknowledge that synthetic material causes real-world harm is a welcome step, but this blueprint is fundamentally a policy wishlist, not an enforceable federal statute.

We think this is a shrewd defensive play by OpenAI. Facing wrongful death lawsuits and mounting pressure from state attorneys general, OpenAI is inviting the prosecutors into the design room. By co-authoring the playbook, OpenAI helps shape the regulatory standard before lawmakers force heavier restrictions on them.

The real test will be accountability. As AG Jackson and Brown noted, voluntary guidelines depend on whether companies actually hold themselves to strict audit trails. If fringe open-source models continue to pump out abuse material without guardrails, OpenAI’s blueprint will protect its own liability while doing little to stop the broader ecosystem's worst offenders.