美国参议员提案禁止超级智能并设立AI部
Sen. Bernie Sanders and Rep. Greg Casar, are unveiling legislation to ban artifi…
推荐理由
美国重量级政客推出针对AI终极风险的硬核监管提案，涉及算力阈值、强制销毁与企业死刑，直接冲击行业合规预期，值得密切关注后续立法进展。
Sen. Bernie Sanders and Rep. Greg Casar, are unveiling legislation to ban artificial superintelligence and create Department of AI
参议员伯尼·桑德斯和众议员格雷格·卡萨尔正在公布一项立法，旨在禁止人工超级智能并设立人工智能部
some elements from the proposal.
提案中的一些要素。
- immediately pause training and deployment of "advanced AI," defined in the bill as systems trained with at least 10^25 operations, until a new federal Department of Artificial Intelligence creates safety rules and model-review processes.
- prohibit systems with "superintelligence precursor" capabilities, including automating AI research, resisting shutdown, self-modifying, evading human oversight, unauthorized hacking, or assisting with nuclear, chemical, or biological weapons.
- give the new department power to force flagged models offline and render them inoperative, while confirmed superintelligence must be destroyed immediately.
- require federal charters for advanced AI developers, including government access to their systems, processes, staff, and physical infrastructure.
- impose what Sanders' office calls a "corporate death penalty" on violating entities, with asset and IP forfeiture under the bill, while certain individuals could face up to 20 years in prison.
- make it US policy to stop superintelligence globally through international agreements, allied coordination, and potentially AI-compute export controls.
- 立即暂停训练和部署“先进人工智能”，该法案将其定义为至少使用 10^25 次运算进行训练的系统，直到新成立的人工智能联邦部门制定安全规则和模型审查流程。
- 禁止具备“超级智能先驱”能力的系统，包括自动化人工智能研究、抵抗关机、自我修改、规避人类监督、未经授权的黑客攻击，或协助制造核武器、化学武器或生物武器。
- 赋予新部门权力，强制将标记的模型下线并使其无法运行，同时已确认的超级智能必须立即销毁。
- 要求先进人工智能开发者获得联邦特许，包括政府对其系统、流程、人员和物理基础设施的访问权。
- 对违规实体实施桑德斯办公室所称的“企业死刑”，根据法案没收资产和知识产权，某些个人可能面临最高 20 年的监禁。
- 通过国际协议、盟友协调以及潜在的人工智能算力出口管制，使美国政策在全球范围内阻止超级智能的发展。
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力