California AI Law Vetoed: Is Winter Coming for Artificial Intelligence?

The headlines in the legal sphere have been dominated by developments in California’s approach to artificial intelligence legislation with an AI law that was vetoed and copyright related obligations that were introduced.

The Vetoed California Artificial Intelligence Bill

In an unexpected move, Governor Gavin Newsom vetoed Senate Bill 1047, a proposed law that aimed to impose safety testing requirements on large language models costing over $100 million to train. The bill also sought to establish legal liabilities for any harms resulting from these models.

The bill faced staunch opposition from major AI corporations, as well as a slew of venture capitalists and industry insiders. Despite the bill being significantly diluted from its original form, Governor Newsom dismissed it on Sunday, citing concerns that it was too feeble—a justification that many observers find dubious.

The veto has ignited extensive discussions, with a consensus among commentators that a revised version of the bill may resurface, potentially as soon as next year.

Is This the Onset of the ‘Brussels Effect’ Triggered by the EU AI Act?

The European Union’s AI Act mandates that high-risk AI systems undergo testing to identify suitable and targeted risk management measures. These tests ensure that the AI systems consistently perform their intended functions and comply with the Act’s requirements. Notably, this provision is risk-based and applies exclusively to high-risk AI systems.

It’s uncertain whether these tests will burden AI providers with costs exceeding $100 million. However, it’s clear that the EU’s regulations are more limited in scope compared to California’s new AI proposals. If Governor Newsom follows through on his intentions, the California law could have a broader reach and impose even stricter obligations on big tech companies than the EU AI Act does.

California new copyright related AI obligations

Adding another layer to the regulatory landscape, Governor Newsom signed into law AB 2013. This legislation requires developers to disclose information on their websites about the data used to train their AI systems.

The law broadly defines “generative artificial intelligence” as AI capable of producing synthetic content—such as text, images, video, and audio—that emulates the structure and characteristics of its training data. Applicable to generative AI released on or after January 1, 2022, developers are required to comply by January 1, 2026.

Under this law, developers who make generative AI systems publicly available to Californians must publish detailed documentation about the training data on their websites. The required information includes:

1. Dataset Sources or Owners: Identification of where the data originated.
2. Purpose Alignment: Explanation of how the datasets contribute to the AI system’s intended functions.
3. Dataset Size: The number of data points, presented in general ranges or estimates for dynamic datasets.
4. Data Point Types: Descriptions of the kinds of data included (e.g., label types or general characteristics).
5. Intellectual Property Status: Whether the datasets contain copyrighted, trademarked, or patented material, or if they are entirely public domain.
6. Acquisition Method: Indication of whether the datasets were purchased or licensed.
7. Personal Information Inclusion: Whether the datasets contain “personal information” or “aggregate consumer information” as defined by the California Consumer Privacy Act.
8. Data Processing Details: Information on any cleaning, processing, or modification of the data and the purposes behind these actions.
9. Data Collection Period: The timeframe during which the data was gathered, including notices if data collection is ongoing.
10. Initial Usage Dates: When the datasets were first utilized in developing the AI system.
11. Synthetic Data Use: Disclosure of any synthetic data generation used in development, including its functional purpose.

In contrast, the EU AI Act requires providers placing general-purpose AI models on the European market to:

1. Copyright Compliance Policy: Implement policies ensuring adherence to EU copyright and related laws.
2. Training Data Summary: Publish a sufficiently detailed summary of the content used for training the AI model.

Clearly, California’s requirements are more extensive, potentially posing greater compliance challenges for AI developers.

The overarching hope is that regulators worldwide, particularly in the EU and the US, will harmonize their approaches to AI regulation. Without a consistent framework, businesses may face prohibitive costs trying to navigate and comply with disparate local laws. Initiatives like the EU AI Pact, which promotes global self-regulatory principles, are steps toward such harmonization. Proactive regulation is preferable, as reactive measures following inevitable incidents could lead to overly stringent obligations driven by immediate pressures.

What Are Your Thoughts? Read more on AI law topics HERE and request a demo of our PRISCA AI Compliance whose video of presentation is HERE.