内容来源：https://techcrunch.com/2025/09/13/california-lawmakers-pass-ai-safety-bill-sb-53-but-newsom-could-still-veto/

内容总结：

美国加利福尼亚州参议院于当地时间周六凌晨正式通过一项人工智能安全法案，要求大型科技企业提高人工智能安全协议的透明度。该法案由州参议员斯科特·维纳提出，除强制要求大型AI实验室公开安全标准外，还设立了员工举报人保护机制，并创建名为"CalCompute"的公共云计算平台以提升算力资源可及性。

法案目前已提交至州长加文·纽森处等待签署。纽森尚未对此表态，但去年他曾否决维纳提出的另一项更严格的安全法案，仅签署了针对深度伪造等具体问题的简化版立法。当时纽森强调需"保护公众免受技术带来的真实威胁"，但认为原法案对大型模型"无论是否应用于高风险环境或涉及敏感数据"均采取统一严格标准欠妥。

值得注意的是，本次法案根据企业规模实行分级披露制度：年收入低于5亿美元的"前沿"AI开发商仅需提交安全概要，而超此门槛的企业须提供详细报告。法案已引发硅谷企业及风投机构的反对，OpenAI虽未直接提及该法案，但致信州长主张"只要符合联邦或欧洲标准即应视为合规"。相反，Anthropic公司公开支持该法案，其联合创始人表示在缺乏联邦标准的情况下，"此项立法为AI治理提供了不可忽视的蓝图"。

与此同时，安德森·霍洛维茨基金高管警告称加州等多地AI法案可能"逾越宪法对州际商业监管的限制"，该机构创始人也曾以科技监管为由公开支持特朗普竞选。

中文翻译：

加利福尼亚州参议院于周六凌晨最终通过了一项重要的人工智能安全法案，该法案对大型科技公司设定了新的透明度要求。提案人州参议员斯科特·维纳表示，SB 53法案"要求大型人工智能实验室公开其安全协议，为AI实验室员工建立举报人保护机制，并创建公共云计算平台以扩大算力访问权限（CalCompute项目）"。

该法案现已提交至州长加文·纽森处等待签署或否决。纽森尚未公开评论SB 53法案，但去年他曾否决过由维纳提出的另一项更全面的安全法案，同时签署了针对深度伪造等问题的针对性立法。当时纽森承认"保护公众免受该技术真实威胁"的重要性，但批评维纳先前提出的法案对大型模型实施"严格标准"，而未区分其是否"部署在高风险环境、涉及关键决策或使用敏感数据"。

维纳表示，新法案吸收了州长否决旧法案后召集的AI专家政策小组的建议。据Politico报道，SB 53法案近期修订后规定：年收入低于5亿美元的"前沿"AI模型开发商仅需披露高层级安全细节，而超过该门槛的企业需提交更详细报告。

该法案已遭到硅谷多家企业、风投机构和游说团体反对。OpenAI在致州长信函中虽未直接提及SB 53，但主张为避免"重复监管与标准冲突"，符合联邦或欧洲标准的企业应被视为满足全州安全要求。

安德森·霍洛维茨基金AI政策负责人兼首席法务官近期声称"加州、纽约等地的多项州级AI法案可能逾越宪法对州际商业监管权限的限制"。该基金联合创始人此前曾将科技监管列为其支持特朗普竞选连任的因素之一，特朗普团队随后呼吁实行十年期的州级AI监管禁令。

与此同时，Anthropic公司公开支持SB 53法案。联合创始人杰克·克拉克表示："我们始终主张更倾向联邦标准，但在缺乏联邦立法的情况下，该法案提供了不容忽视的AI治理蓝图。"

英文来源：

California’s state senate gave final approval early on Saturday morning to a major AI safety bill setting new transparency requirements on large companies.
As described by its author, state senator Scott Wiener, SB 53 “requires large AI labs to be transparent about their safety protocols, creates whistleblower protections for [employees] at AI labs & creates a public cloud to expand compute access (CalCompute).”
The bill now goes to California Governor Gavin Newsom to sign or veto. He has not commented publicly on SB 53, but last year, he vetoed a more expansive safety bill also authored by Wiener, while signing narrower legislation targeting issues like deepfakes.
At the time, Newsom acknowledged the importance of “protecting the public from real threats posed by this technology,” but criticized Wiener’s previous bill for applying “stringent standards” to large models regardless of whether they were “deployed in high-risk environments, [involved] critical decision-making or the use of sensitive data.”
Wiener said the new bill was influenced by recommendations from a policy panel of AI experts that Newsom convened after his veto.
Politico also reports that SB 53 was recently amended so that companies developing “frontier” AI models while bringing in less than $500 million in annual revenue will only need to disclose high level safety details, while companies making more than that will need to provide more detailed reports.
The bill has been opposed by a number of Silicon Valley companies, VC firms, and lobbying groups. In a recent letter to Newsom, OpenAI did not mention SB 53 specifically but argued that to avoid “duplication and inconsistencies,” companies should be considered compliant with statewide safety rules as long as they meet federal or European standards.
Join 10k+ tech and VC leaders for growth and connections at Disrupt 2025
Netflix, Box, a16z, ElevenLabs, Wayve, Sequoia Capital, Elad Gil — just some of the 250+ heavy hitters leading 200+ sessions designed to deliver the insights that fuel startup growth and sharpen your edge. Don’t miss the 20th anniversary of TechCrunch, and a chance to learn from the top voices in tech. Grab your ticket before Sept 26 to save up to $668.
Join 10k+ tech and VC leaders for growth and connections at Disrupt 2025
Netflix, Box, a16z, ElevenLabs, Wayve, Sequoia Capital, Elad Gil — just some of the 250+ heavy hitters leading 200+ sessions designed to deliver the insights that fuel startup growth and sharpen your edge. Don’t miss the 20th anniversary of TechCrunch, and a chance to learn from the top voices in tech. Grab your ticket before Sept 26 to save up to $668.
And Andreessen Horowitz’s head of AI policy and chief legal officer recently claimed that ”many of today’s state AI bills — like proposals in California and New York — risk” crossing a line by violating constitutional limits on how states can regulate interstate commerce.
a16z’s co-founders had previously pointed to tech regulation as one of the factors leading them to back Donald Trump’s bid for a second term. The Trump administration and its allies subsequently called for a 10-year ban on state AI regulation.
Anthropic, meanwhile, has come out in favor of SB 53.
“We have long said we would prefer a federal standard,” said Anthropic co-founder Jack Clark in a post. “But in the absence of that this creates a solid blueprint for AI governance that cannot be ignored.”