OpenAI Forum: Using AI to Protect Children Online

In a recent OpenAI Forum conversation, Julie Cordua and OpenAI’s Chelsea Carlson discussed how the scale and complexity of online harm continue to grow and have outpaced purely human moderation. Emerging forms of abuse, from AI-generated content to rapidly evolving grooming and sextortion tactics, require detection systems that can adapt as quickly as platforms innovate.

Over the course of the conversation, Julie spoke on five key takeaways that technology platforms are dealing with today:

1. The nature of child sexual exploitation has fundamentally shifted with technology
   Today, harms like grooming, sextortion, deepfake nudes, and coordinated online exploitation can originate entirely online, making the problem more complex and scalable than ever before.
2. AI is essential — but not exhaustive — in combating abuse at scale
   Given the massive volume of online content, hash matching and AI classifiers are critical for detecting both known and new abusive material. However, effective protection requires a combination of AI, human expertise, platform design, policy, and education.
3. Detection and response must operate with massive scale and speed
   AI classifiers help triage vast datasets (billions of files or terabytes of seized data) to quickly identify high-risk cases, enabling faster intervention and potentially saving children from ongoing harm.
4. Prevention requires a layered, ecosystem-wide approach
   Meaningful progress depends on multiple coordinated efforts. No single intervention works alone, as this is a systems problem.
5. Human impact is a critical constraint
   Training models and investigating cases often requires humans to review deeply distressing content, leading to burnout and limited workforce capacity.

Watch the full conversation to learn more about the state of the child safety ecosystem and the challenges Thorn helps solve every day.