AI安全公司Anthropic向美国国会提议了一份AI安全法案草案 1。该提案规定，当超过50人面临死亡风险或可能造成10亿美元以上财产损失时，且风险源于生物武器创造、模型逃逸开发者控制或网络攻击这三类灾难性情景，应阻止相关模型的发布 1。
法案草案建立了一套独立评估机制 1。由美国商务部登记的独立评估机构负责进行风险评估，在发现灾难性风险后有72小时向商务部报告的期限 1。一旦风险被确认，商务部可向特区法官申请停止模型发布，但联邦机构本身不具有直接决定权 1。
Anthropic发言人表示："该法案代表了Anthropic高级AI框架中的提议，我们自6月以来一直在与联邦和州立法者倡导这些提议" 1。该提案的风险阈值与加州法案Senate Bill 53相似 1。
AI safety company Anthropic has submitted a proposed legislative framework to the U.S. Congress that would restrict the release of AI models under specific catastrophic risk scenarios 1. The bill would block model deployment if more than 50 people face potential death or property losses exceeding $1 billion could result, provided the risks stem from bioweapon creation, model escape from developer control, or cyberattacks 1.
Under the proposal, independent assessment bodies registered with the U.S. Department of Commerce would evaluate AI models for these catastrophic risks 1. Such evaluators would have 72 hours to report identified dangers to the Commerce Department, which could then petition a federal judge to halt model release 1. Notably, the framework prevents federal agencies from independently making the decision to block deployment 1.
An Anthropic spokesperson stated that "the bill represents proposals from Anthropic's advanced AI framework, which we have been advocating for with federal and state legislators since June" 1. The risk thresholds in this proposal align with those in California's Senate Bill 53 1. The company's position represents a more interventionist stance than that of Nvidia Chief Executive Jensen Huang, who has argued the AI industry does not require new regulations 1.
评论
还没有评论，欢迎留下第一条。