California’s Democratic governor, Gavin Newsom, vetoed a local Parliament law that aimed to regulate the development of artificial intelligence (AI), criticizing the text for not being sufficiently targeted.
SB 1047 was widely adopted in August by the California Senate and Assembly, the state’s lower house.
Evidence needed to assess AI risks
In particular, it established the need for AI developers to test their models to assess risks before putting them online.
An AI model is software that is fed with an immense amount of data to predict the best response to a query.
It is, in particular, the basis of so-called generative AI, which can generate a text, an image, a video or a sound following a request in everyday language.
The text, presented by the local Democratic senator from San Francisco Scott Wiener, also obliged these technology companies to equip the model with a circuit breaker, which would allow it to be put into sleep mode in the event of a slip or proven risk.
The law also established the principle of corporate liability in the event of serious damage caused by a model, with the possibility of prosecution by the California prosecutor.
Private AI players oppose the law
Several private AI players had publicly expressed their opposition to this text, warning that it risked driving tech companies and investors out of California.
But others, researchers and companies, welcomed it, believing it necessary to strictly regulate the emergence of AI.
According to the governor of California, 32 of the 50 largest companies in the AI sector are located in the “Golden State,” as the state is known.
Gavin Newsom maintained suspense until the last moment regarding his position on this law.
In a letter dated Sunday, he explained the reasons for his veto, considering in particular that the text only applies to “the most expensive and largest models”, while “smaller models could be more dangerous.”
Another complaint: Senator Wiener’s formulation “encompasses all functions, even the most ordinary ones”, that AI can perform, while it should focus on models “deployed in high-risk environments”, “that involve critical decision-making” or “the use of sensitive data.”
To demonstrate his desire to control the development of AI, Gavin Newsom recalled that during the last month he enacted seventeen laws in this direction.
They reinforce transparency obligations towards the public in relation to the use of AI and limit, among other things, “deepfakes”: photographs, videos and sounds that give the impression of being authentic when they were created from scratch.
Source: BFM TV
