OpenAI Backs California’s Groundbreaking Bill to Tag AI-Generated Content

OpenAI supports a California bill that requires provenance data for AI-generated content to combat deepfakes, but it opposes a separate bill that imposes strict AI development regulations.

Microsoft-backed OpenAI has endorsed a bill currently advancing through the California state legislature. The bill’s purpose is to prevent individuals from misusing AI-generated content. Assemblymember Buffy Wicks of District 14 introduced the bill, AB 3211, to widespread support. The Assembly has already approved it with a vote of 62-0. Before the end of the month, the state senate should vote on the bill.

The California Digital Content Provenance Standards bill says that gen-AI providers must use provenance data on synthetic content created or significantly changed by a generative AI system that the provider makes available, as those terms are defined, and must do adversarial testing exercises as per the bill summary. Reuters says that OpenAI’s Chief Strategy Officer, Jason Kwon, wrote Wicks a letter supporting the bill.

“New technology and standards can help people understand the origin of content they find online and avoid confusion between human-generated and photorealistic AI-generated content,” Kwon wrote in the letter. It also says that big websites like social media sites that “can disseminate specified content must use labels to disclose, as specified, any machine-readable provenance data detected in synthetic content that is distributed on its platform.”

Tech Industry Divided on California’s SB 1047

The majority of individuals in the tech industry continue to endorse the aforementioned watermarking bill, aimed at curbing the proliferation of AI deepfakes. However, OpenAI and others oppose SB 1047, which the state Senate passed by a vote of 32-1 earlier this month.

State Senator Scott Wiener of District 11 authored SB 1047, which would significantly alter the development and application of AI in California. It’d “require that a developer comply with various requirements before beginning to initially train a covered model, as defined, including implementing the capability to promptly enact a full shutdown, as defined, and implement a written and separate safety and security protocol, as specified.”

If a company doesn’t follow the rules, the bill would let the California Attorney General sue them in civil court. As part of the proposed bill, the Government Operations Agency would get a Board of Frontier Models to help keep an eye on it. Many companies and groups oppose SB 1047. Some examples are Andreessen Horowitz, Meta Platforms, UC Berkeley, and many AI startups.