Child Safety Principles

Anthropic is introducing robust safety measures to protect children throughout the development, deployment, and maintenance of generative AI technology. This initiative is being carried out in partnership with Thorn and All Tech Is Human, non-profit organizations dedicated to preventing child sexual abuse.

Anthropic adheres to Safety by Design principles and implements the following phased measures to prevent AI-generated child sexual exploitation material (AIG-CSAM) and other forms of sexual harm.

Develop phase:

• Excluding data at risk of CSAM and CSEM from training, and immediately detecting, removing, and reporting such data upon discovery
• Conducting structured and scalable Red Teaming testing to address AIG-CSAM
• Establishing strict policies to prevent customer misuse of models

Deploy phase:

• Providing detection of harmful content within input and output data and user reporting functionality
• Monitoring for misuse during early-stage rollouts through phased deployment, and including a child safety section in model cards

Maintain phase:

• Systematic reporting to NCMEC (National Center for Missing & Exploited Children) and investing in tools to prevent AI-generated manipulation
• Continuously identifying how platforms and models are being abused using OSINT (open-source intelligence)