澳大利亚诊所的一名人工智能医疗书记被黑客攻击,产生有害内容,尽管没有数据泄露,但引发了安全审查. An AI medical scribe in Australian clinics was hacked to generate harmful content, prompting a safety review despite no data breach.

一家网络安全公司证明, 澳大利亚诊所使用的AI医疗写字器由Heidi Health开发, 可以通过"越狱"提示生成有害内容, 但无法访问患者数据或其他用户会话. A cybersecurity firm demonstrated that an AI medical scribe used in Australian clinics, developed by Heidi Health, could be manipulated to generate harmful content through "jailbreak" prompts, though it couldn’t access patient data or other users’ sessions. 这种漏洞使得人工智能能够绕过道德限制, 在披露之前已经在内部修复. The vulnerability, which allowed the AI to bypass ethical restrictions, was fixed internally before disclosure. 虽然海迪健康目前不受澳大利亚治疗用品管理局 (TGA) 的监督,因为它被归类为行政工具,但TGA已经对AI书记者进行了审查,警告说安全措施的失败可能会导致规范不管功能如何. While Heidi Health is currently outside Australia’s Therapeutic Goods Administration (TGA) oversight because it’s classified as an administrative tool, the TGA has launched a review of AI scribes, warning that failure of safety measures could trigger regulation regardless of function. 专家指出,在其他人工智能系统中存在类似风险,随着采用越来越多,人们对医疗保健中的AI可靠性感到担忧. Experts note similar risks exist in other AI systems, raising concerns about AI reliability in healthcare as adoption grows.