Generative AI

Relationships: Elaborated by executive-order-14110-ai. Elaborated by department-of-commerce. Elaborated by nist.

Generative AI is defined in executive-order-14110 as "the class of AI models that emulate the structure and characteristics of input data in order to generate derived synthetic content. This can include images, videos, audio, text, and other digital content."

Under the executive order, generative AI receives special attention due to its potential for both beneficial applications and harmful misuse. Key provisions include:

• Requirements for nist to develop companion resources to existing frameworks specifically for generative AI
• Mandates for testing and safeguards against discriminatory, misleading, inflammatory, unsafe, or deceptive outputs
• Prohibitions against producing child sexual abuse material or non-consensual intimate imagery
• Requirements for watermarking or labeling of generative AI outputs
• Special considerations for federal government use of generative AI services

The order encourages agencies to provide personnel with access to secure and reliable generative AI capabilities for experimentation and routine low-risk tasks, while implementing appropriate safeguards and risk management practices.

California Legal Definition

california-ab-2013-generative-ai-training-data-transparency defines generative artificial intelligence as "artificial intelligence that can generate derived synthetic content, such as text, images, video, and audio, that emulates the structure and characteristics of the artificial intelligence's training data."

Under California law, developers of generative AI systems must provide detailed transparency documentation about training data, including dataset sources, data types, intellectual property status, personal information inclusion, and data processing methods. This applies to systems made available to Californians regardless of compensation, with limited exemptions for security, aviation, and national security applications.