AI transparency rule from Bay Area lawmakers kicks in Saturday

After several years of delays, legislation from two Bay Area lawmakers that is being called a landmark artificial intelligence transparency law by academics and advocates is set to go into effect Aug. 1.

First passed in 2024, the law will require AI-generated content like images and videos to contain easily accessible embedded details about its origin and history that will help users to identify content that is AI-generated, along with providing users a free AI detection tool.

“Three years ago, the problem hit me in the face,” said state Sen. Josh Becker, D-San Mateo, one of the co-authors of the legislation, along with Assemblymember Buffy Wicks, D-Oakland, and Rick Chavez Zbur, D-Los Angeles.

Becker said learning of AI scams surrounding deep fakes and election misinformation galvanized him to take action.

“I realized we kind of have to come up with a way to let people know was something AI or not,” he said.

Metadata is a part of all digital content, often used to describe various aspects such as an author of a book or the name of a song. This law will require AI developers to automatically apply when and where a piece of content was created once the AI content is generated. This information will be attached to the piece of content into the future, so viewers can always trace back its point of origin.

According to Becker, embedding origin details into the metadata of AI content will be harder to fake as opposed to a watermark, or a visual indicator that an image or video was generated with AI, which could easily be added to real content or removed from AI-generated content.

While this effort marks a step forward in transparency, Becker said the work will be ongoing as technology develops and changes. For now, Becker said this law is the cumulation of three years of effort and work to figure out what will be most effective.

California decided to delay implementation of the transparency law to align with the European Union’s AI Act, which takes effect Aug. 2. While Europe’s regulation touches various aspects of AI regulation, California’s transparency requirements largely line up with Europe.

Becker added that enforcement will be a key part of ensuring the legislation is effective, noting that violators will be liable for a civil penalty in the amount of $5,000 per violation, which can be litigated through a civil action filed by the state’s attorney general, a city attorney or county council.

The next phase of the law’s rollout will occur in January, when large social media companies will be required to detect and disclose this data for the content they distribute, while also providing users with an interface to show if the content was AI-generated.

This story was written by Eric Urbach for Bay City News Service.