In California, a center for AI technology development, the brakes have been applied to a controversial AI regulation bill. California Governor Gavin Newsom announced on the 29th that he had vetoed the AI regulation bill ‘SB 1047’, which strengthens the responsibility of AI development companies. It was just over a month after the state legislature passed the bill on the 28th of last month, and the veto was exercised one day before the signing deadline on the 30th.
Governor Newsom explained the reason for his veto, saying, “The regulations are focused only on the largest and most expensive AI models,” and “It only tried to regulate based on the size and cost of AI models, without considering whether the models are actually being used in dangerous situations.” He continued, “Even small AI models can cause very dangerous situations that handle sensitive data such as the power grid or medical records, while large models are used for relatively low-risk tasks such as customer service.” He added, “I agree with the bill’s supporters that we need to prepare for a major disaster before it happens, but regulations must be based on scientific and empirical evidence.”
The veto by Governor Gavin has given a breather to technology companies that have opposed the bill a chance to take a breather. The bill requires safety testing for large-scale language models costing more than $100 million and requires developers to take reasonable care to ensure that AI systems do not cause serious problems such as multiple deaths or $500 million in property damage.
It also introduces a “kill switch” to stop AI when it becomes difficult to control and allows the state attorney general to sue if regulations are not followed. The bill was introduced by Democratic state senator Scott Wiener, who said, “We need to create laws to protect the public before AI becomes uncontrollable,” and passed the assembly. However, the tech industry, including Microsoft (MS), Meta, and ChatGPT developer OpenAI, opposed the bill, saying that it would slow down technological innovation. The controversy grew as former House Speaker Nancy Pelosi and San Francisco Mayor London Breed also expressed their opposition.
Governor Newsom emphasized that he would conduct an analysis of the risks and capabilities of major AI models led by several scholars, including AI scholar and entrepreneur Fei-Fei Li, along with announcing his veto of ‘SB 1047’. Meanwhile, Governor Newsom signed and promulgated a privacy law amendment on the 28th to prevent people’s brain information from being misused by neurotechnology companies. According to the New York Times (NYT), the bill protects ‘neural data’ such as brains in the same way as other information that is already considered sensitive data under California’s Consumer Privacy Protection Act, such as biometric information such as facial images, genes (DNA), and fingerprints.
As developments are underway to develop devices that help address cognitive issues, such as meditation, improved concentration, and treatment of mental health conditions such as depression, there is a risk that personal brain information could be misused. “Neural data” also includes data generated by the peripheral nervous system, which extends from the brain and spinal cord to the rest of the body. As it is included in the scope of personal information that must be protected, people can request, request deletion or correction of their own data collected by neurotech companies. They can also request that companies that sell or share their neural data exclude their information.
“This bill is critical to protecting the privacy of neural data that belongs to people,” said Senator Josh Becker, a California Democrat who sponsored the bill. Currently, not only startups but also large information and communication companies such as Facebook parent company Meta and Apple are developing devices that can collect massive amounts of brain and neural information.
TechNet, a lobbying group representing these companies, has opposed the bill, arguing that it would regulate almost all technologies that record human behavior. TechNet’s argument was not accepted, but the New York Times reported that the bill does not apply to information inferred from non-neural data of the human body, such as heart rate or blood pressure.