OpenAI Child Safety Commitment: Adopting Safety by Design Principles

OpenAI, in collaboration with industry leaders including Amazon, Anthropic, Civitai, Google, Meta, Metaphysic, Microsoft, Mistral AI, and Stability AI, has adopted the Safety by Design principles. This initiative, led by the nonprofit Thorn and the organization All Tech Is Human, aims to proactively mitigate the risks that generative AI poses to children, ensuring child safety is prioritized throughout the development, deployment, and maintenance of AI technologies.

Proactive Development of Child-Safe AI

OpenAI commits to building and training generative AI models that proactively address child safety risks. This development phase focuses on three primary technical and operational goals:

  • Training Data Integrity: OpenAI will responsibly source training datasets and actively detect and remove child sexual abuse material (CSAM) and child sexual exploitation material (CSEM) from training data, reporting any confirmed CSAM to the relevant authorities.
  • Iterative Testing: The development process incorporates feedback loops and iterative stress-testing strategies to identify vulnerabilities.
  • Adversarial Misuse: The company is deploying solutions specifically designed to address and prevent adversarial misuse of the models.

Deployment and Distribution Protections

To ensure safety during the release and distribution of generative AI models, OpenAI has committed to the following measures:

  • Evaluation and Prevention: Models are released and distributed only after they have been trained and evaluated for child safety, with protections integrated throughout the process.
  • Abuse Response: OpenAI will combat and respond to abusive content and conduct, incorporating prevention efforts to limit harm.
  • Developer Ownership: The company encourages developer ownership in the following Safety by Design principles.

Ongoing Maintenance and Risk Mitigation

OpenAI maintains platform safety by continuously monitoring and responding to emerging child safety risks. Key commitments in this commitment include:

  • Combating AIG-CSAM: OpenAI is committed to removing AI-generated child sexual abuse material (AIG-CSAM) generated by bad actors from its platforms.
  • Research Investment: The company will invest in research and future technology solutions to combat CSAM, AIG-CSAM, and CSEM.
  • Stakeholder Engagement: OpenAI continues to engage with the National Center for Missing and Exploited Children (NCMEC), the Tech Coalition, and the company's government and industry stakeholders to enhance reporting mechanisms and child protection issues.

Accountability and Reporting

As part of this working group, OpenAI and its peers have agreed to release progress updates every year to maintain transparency and accountability regarding their implementation of these safety principles.

Sources