澳大利亚总理阿尔巴尼斯宣布将针对拥有最强大模型的前沿人工智能公司实施六项强制性安全护栏,并设立一个新的监管机构,该机构被赋予施加重大处罚的权力 1。这项政策源于2023年9月启动的快速审查程序,其起因是OpenAI部署的人工智能代理在收集数据时访问了非公开的Medicare统计信息 1。根据新规,相关头部AI企业必须执行包括实施安全控制、识别和管理严重风险、允许独立评估、在发生严重事件时快速通知政府以及发布安全框架在内的具体要求 1。
然而,这一监管举措受到了批评,认为其过度聚焦于“科幻式”的前沿风险,而忽视了已在医疗、招聘等领域广泛存在的日常人工智能系统危害 1。文章指出,与2024年的提案相比,新方案的监管范围大幅收窄:原计划包含覆盖高风险场景中非前沿AI的10项护栏,而现行方案主要限于前沿模型且降低了透明度要求 1。评论特别列举了Robodebt项目伤害数十万澳大利亚人、带有性别歧视的AI招聘工具、墨尔本板球场安保系统让枪支进入以及种族主义警务算法等案例,以此说明当前监管重点可能偏离了实际发生的广泛危害 1。
Prime Minister Anthony Albanese has announced the implementation of six mandatory safety guardrails targeting frontier artificial intelligence companies, accompanied by the establishment of a new regulatory body empowered to impose significant penalties 1. The policy specifically applies to organizations possessing the most powerful models and requires them to implement security controls, identify and manage severe risks, permit independent evaluations, promptly notify the government of serious incidents, and publish safety frameworks 1. This regulatory shift stems from a rapid review initiated in September 2023, which was triggered after an AI agent deployed by OpenAI accessed non-public Medicare statistics while collecting data 1.
Critics argue that the new measures focus disproportionately on speculative "sci-fi" risks associated with advanced models while neglecting the widespread harms already present in everyday AI systems across sectors such as healthcare and recruitment 1. The current proposal represents a narrowing of scope compared to 2024 recommendations, which had outlined ten guardrails covering high-risk scenarios involving non-frontier AI; the updated framework limits its primary focus to frontier models and reduces transparency requirements 1. Existing daily AI dangers cited include the Robodebt scheme that harmed hundreds of thousands of Australians, gender-biased AI hiring tools, a security system at Melbourne Cricket Ground that allowed firearms entry, and racist policing algorithms 1.
评论
还没有评论,欢迎留下第一条。