Vintage illustration of the California Council. sauce
Along with the New Year, a new legislative meeting will appear in a US state house. As the policy creator returns to work, the promise and danger of artificial intelligence (AI) remains the spotlight. Last year, the 700 AI bill, which was introduced nationwide, created a new state law bumper crop, and the momentum did not show signs of deceleration.
When evaluating their legal priority, the State Council has taken the stock of the checkbox they have carved last year. The punch list, a legal target policy policy, is drawn out of the steady infusion of a report issued or soon issued by the AI Task Force in the lower house of Illinois and Virginia.
As reflected in these reports, AI affects cyber security, data protection, energy and environment, the freedom of citizens, and the provision of public services. Given the wide impact, the members of the Diet are still cutting out their work even after last year’s achievements. Topics that prioritize the priority are a traver that requires a politician who is familiar with potential art in order to evaluate political will, the members of the members, and the urgent real world. Evidence of rapid evolution of social, technical, economic and global political development.
For example, a politician banning the sexual deep fake of children, the topic of more than the last 20 state law, or a problem that the government regulates, or the government regulates it. There may be some simple victory on the problem. When establishing a new state office or an advisory committee in AI, itself itself. Picking up relatively low feather fruits is optically important for members. This is because they can introduce those victories to the general public when navigating strictly legislative issues. These are more than one of the most commonly used technical expertise in caution, guide the constitutional needles (as described below), assign executive institutions, or are nervous with each other. Includes a topic corresponding to stocks (corresponds to multiple shares that may have a large lobby activity behind it).
Beyond domain -specific bills, some councils increase their vision. 2025 can be expected to bring various efforts to regulate the AI warrant notified by the European Union and the new Landmark law. In Colorado, it is the first state in Japan that has successfully passed the comprehensive AI bill. In some states, they have not lost their time. At the end of last year, Texas members have introduced a major AI bill that has already been criticized for being too far and not far away.
Lessons learned from California
In California, where many famous AI companies are at the headquarters, the Diet set up themselves after the 2024 multiple sessions. Last year, Governor Gavin Newsom signed 17 AI -related bills that cover a wide range of policy domains, from education to labor, privacy, health care, and elections. In addition, the governor refused to SB 1047. This is a comprehensive comprehensive negative effect on (especially) the potential negative effect on innovation and the potential negative impact on the use of the power and cost of certain model training computing. This is an AI safety bill.
SB 1047 and its end MISE have attracted more media than all AI bills signed by all AI bills governor Newsum. As the governor’s office emphasized in a public message, it is a pity that it gives the government’s incorrect impression when the last legislation session is actually produced significantly due to AI issues. is. California’s new law raft suggests that policy proprietors should not index over indexes to compare the fate of the comprehensive AI bills in California and Colorado. 。 The parliamentarians have the same goal of whether to concentrate energy on major European -style laws, which are trying to cover a large amount of foundation with a single bill, or instead of specific use cases and bills that have dealt with harm. It is necessary to carefully consider whether it can be achieved. It may also prove that it is potentially politically executed.
It doesn’t mean that a comprehensive AI method is not worth trying, but that you should not replace the more progressive problem with a more targeted problem. For example, in addition to a comprehensive AI bill, Texas Council has also introduced several individual bills for harmful AI, such as malicious and sexual deep fake.
One advantage of the bill package for the “one bill to control all” is that a narrow local bill can be clearly dismantled than the same language is sandwiched between larger laws. The controversy of the complex bills can create enough drags to prevent the legislative session from expiring. Similarly, if a court attacks a part of the law enacted, the entire law may be at risk if it is not carefully removed. Some use of AI’s initial efforts to regulate AI’s various deceptive CEPT or harmful use are unstable, so some use cases regulations fall when others fall. You may be able to withstand the scrutiny of the judiciary. By cutting out a part of the AI method patchwork quilt, the rest still remains in a predetermined position and you can work as intended.
A timely case is a state law that deals with a deceptive depth related to election, and two similar laws have encountered different fate in court. In early October 2024, the Federal District Court banned one of the three new laws in California on this topic on the basis of revision, just two weeks after the bill 2839 became the law. (This ruling was probably hidden by SB 1047 veto early the week, so it seems that almost no notifications other than the high -tech policy circle were attracted.)
A strong opinion stated in words that the court did not reach the strict standards required for AB 2839 for speech -based content -based content to survive the constitution. Even the deceived or false speech is generally estimated by Article 1 of the Constitution (but for a small, established exception), so it is “substantially deceived” in the election. The California law, which regulates deep fake, clearly had a high bar. The court has refused to be characterized as an attempt to “buldou over” for long -standing freedom protection. The court has banned everything, except for one small provisions of laws and regulations. The court has temporarily continued to execute another election visit law AB 2655, and its constitutionality is also a problem. (Several cases that challenge both or both laws are integrated in the original case.)
These developments are in contrast to the opinions just issued on January 10 by the Federal Court in Minnesota, and by the same plaintiffs as the conservative political content creator, the original California case. The same litigation was submitted to the opposite result. The court rejected the plaintiff’s demands to prohibit the Deep Face Law in the Minnesota election in advance, and inferred that he had labeled his deep -fark ten as a parody (a speech protected by the Constitution). The court stated that the disclosure had left his content outside the law. In contrast to the intense PAEAN for the first revision of the California court, the Minnesota Court is deeply controlled by important constitutional questions (such as tolerance for compulsory labeling of synthesis) by law. I avoided doing things neatly.
Councilors need to pay close attention to these cases. Limiting the content that is generated or modified by AI is a difficult business, especially in consideration of the rugged constitutional protection against American speeches, especially in the core of the first correction. Nevertheless, 20 states have already enacted political deep -fur. Since the November election is behind us, the rest may feel that there is little urgency about this issue during this year’s off -cycle year.
In 2025, the state fires more rockets before seeing last year’s land.
The legislative efforts to deal with the harm of the AI are free expressions, privacy, consumer protection, individual dignity, safety, economic impact, and innovation, and a lot of profits walk on detailed lines. I have to. Since the actual test of legislative productivity in 2024 has not yet come, it is not yet known whether the AI method has been balanced with any specific AI method. With the delay, the target entity gives the opportunity to enter the compliance, and the regulatory authorities will issue guidance and have time to plan an execution.
Nevertheless, the state is now looking for an example of how to write a unique AI method, without knowing how it actually develops. The dynamics may be inevitable due to the current (actually global) enthusiasm of AI regulations. The combination of the Brussels effect, the desire to see the politicians can achieve things, the very realistic harm of AI, and the fierce pace of technology development are “wait and see” among members. It is hardly patient. Therefore, the 2025 legislative conference is expected to be more prolonged for the introduced AI bill than 2024 than 2024.
As artificial intelligence continues to reach the multifaceted aspects of our lives, the law must deal with the current effects of AI and predict what will come next. Under the control method, policies can benefit from state -of -the -art research from Stanford and other institutions. With the evidence -based policy approach, lawmakers can help them to provide services not only as voters and members, but also as humans.
