Non-consensual intimate imagery (NCII)

Mitigation strategy

1. Prioritized Mitigation: Implement rigorous pre-deployment and continuous runtime 'red teaming' and safety testing protocols to proactively identify and neutralize model vulnerabilities that could allow the generation of non-consensual intimate imagery (NCII). Enforce strict input/output content filtering at the system level to block and log prompts violating terms of service regarding the sexual exploitation of human likeness. 2. Content Provenance and Distribution Interventions: Integrate and enforce the use of content hashing technology (e.g., digital fingerprinting) across all content distribution channels to enable the rapid detection, blocking, and removal of known NCII upon subsequent attempted sharing, effectively curtailing scaled distribution. 3. Organizational and Regulatory Alignment: Establish clear, internal AI governance policies that explicitly prohibit the use of generative AI capabilities for the creation of NCII, ensuring full compliance with evolving state and international legislation that criminalizes the production and distribution of non-consensual deepfakes.