纽约市议会正审议一项 AI 监管法案,要求在本地营销、销售或部署的 AI 模型先通过外部验证,否则供应商每次违规销售或部署将被罚款 $25K。法案还要求模型具备由人类操作员临时或永久停止运行的关停能力,验证方将评估准确性、校准、确定性、延迟、数据来源、偏见、隐私和安全,并向该市 Cyber Command 认证。
Ex-Anthropic researcher Jacob Coxon told New York City lawmakers that humans will probably lose control of AI and is set to testify at NYC AI hearing on ‘kill switches,’
New York City is also prepareing a bill that would require every AI model sold locally to pass outside validation first and would fine AI vendors $25K per unvalidated sale or deployment.
All 51 council members are weighing this lead bill that would outlaw marketing, selling or deploying any AI model in the city without outside validation.
It also requires a shutdown capability, meaning a human operator can technically make the model stop working, temporarily or permanently.
Validators would grade accuracy, calibration, determinism, latency, data provenance, bias, privacy and safety, then certify readiness to the city's Cyber Command.
来源:Rohan Paul · x.com