AI 智能体开始代医务人员填写医疗表格并提交保险申请

Nano Banana Pro 生成,提示词来自 THE DECODER
OpenAI 将 GPT-6 Astra 和 Luna 部署为软件智能体,通过鼠标点击和键盘输入来操作诊所软件与保险门户。
在测试中,这两款模型都能可靠地处理医疗行政任务。GPT-6 Luna 每项任务的成本约为 8.5 美分。
包括 CVS Health 在内的美国公司已经在使用这些智能体处理转诊。该产品目前面向美国市场及其数据隐私要求。
OpenAI 希望 GPT-6 Astra 和 GPT-6 Luna 像数字文员一样,在诊所软件和保险门户中点击、输入来完成工作。据该公司称,GPT-6 Luna 处理行政任务的成本约为每项任务 8.5 美分。
在最近的一次产品发布中,OpenAI 瞄准了医疗行业缺乏 API 的问题。9 月下旬集成到 Agents API 中的计算机使用智能体,会像人类一样操作现有的用户界面。在此场景中,它们在电子健康档案与外部网络门户之间传输数据。
据 OpenAI 介绍,当患者反馈背痛时,该智能体应检索相关的既往检查结果,并起草一份 MRI 检查申请以及对患者的回复。它还应在保险公司的门户网站上准备一份预授权申请。临床医生和负责的行政人员在发布前会审核这些草稿。这些功能可通过 API 使用,也可在 ChatGPT for Healthcare 和 ChatGPT for Clinicians 中使用。
GPT-6 Luna 以每项任务 8.5 美分的成本完成行政任务
在 OpenAI 描述的一项评估中,GPT-6 Astra 处理了来自电子健康记录收件箱的任务,其中包括为一名手臂骨折的儿童准备布洛芬剂量。医生评审员将所评估的 325 项任务中的 99% 评为临床安全。在 646 次评审中,有 93% 的评审认为这些准备工作非常有帮助或极有帮助。
医生以最大推理强度对 GPT-6 Astra 的收件箱任务进行评估。|图片来源:OpenAI
对于事先授权、申诉或医疗设备订购等行政工作流程,OpenAI 指向了斯坦福大学研究人员的独立基准测试 HealthAdminBench。GPT-6 Astra 获得了可能得分的 94.3%。专为低 API 定价而打造的 GPT-6 Luna 得分为 91.8%,据 OpenAI 称,其每项任务成本约为 8.5 美分,比 Astra 低 98%。两个模型在该基准测试中均以最大推理强度运行。
美国主要医疗健康公司测试并部署 OpenAI 的计算机操作智能体
多家合作机构正在帮助推进这项技术。在 CVS Health,这些智能体已经能够跨不同系统处理转诊和专科会诊。根据公告,一个智能体在一个周末内清理了积压数月的待发文件,另一个智能体则每天处理数万次会诊的系统操作步骤。CVS Health 已将其软件架构以 Agent Kit 的形式发布在 GitHub 上。
在保险公司 Cigna Healthcare,这些模型整合来自 10 多个独立系统的患者信息。Abridge 正在开发这些智能体,计划用于临床出院规划。Ambience 展示了额外的安全机制:当患者记录不清晰时,智能体会停止操作并向医生请求输入。像 Assort 这样的初创公司也在 Codex 中使用该功能,在 API 不支持最后一步点击的诊所管理软件中完成转诊。
数据隐私、健康记录与欧洲
由于这些智能体直接在电子健康记录中工作,保护敏感健康数据是核心关切。OpenAI 表示,不会使用组织数据来训练其基础模型。据该公司称,ChatGPT for Healthcare 提供基于角色的访问控制和审计日志。组织还可以配置应用和环境,使智能体仅访问特定任务所需的记录并执行所需操作。
对于涉及受保护健康信息、符合 HIPAA 的工作流程,组织必须与 OpenAI 签署商业伙伴协议(BAA)。它们还需要使用协议覆盖的服务,并应用所需的安全与数据保留设置。医疗机构自身对合规使用负责,但作为签约方,OpenAI 在 HIPAA 下也有自身义务。美国经过验证的医疗专业人员若使用免费个人版 ChatGPT for Clinicians,也可以签署 BAA,不过该协议仅适用于其覆盖的功能。
据 OpenAI 称,在 CVS Health 的部署中,数据保留在 CVS 掌控的基础设施内。CVS 通过截图和操作日志审查智能体的工作步骤。公告未说明这些截图在哪里处理,也未说明它们会被存储多久。公告同样未涉及欧洲《通用数据保护条例》(GDPR)。到目前为止,该产品仅面向美国市场。
OpenAI 主要将这项技术定位为一种工具,用来应对高度碎片化的医疗体系中的行政摩擦。它能否被广泛采用,可能不仅取决于该公司公告中提到的准确率和成本。同样具有决定性的是:智能体与现有系统的连接是否可靠,以及 AI 智能体在实践中能否被可靠地监控。另一个悬而未决的问题是,涉及临床或法律相关步骤时,是否必须由人工来做出决定。
没有炒作的 AI 新闻——由人工精选
译文由英文原文机器翻译,术语按行业惯例统一,可在文末展开原文核对。
查看英文原文(AI agents are starting to fill out medical forms and file insurance requests in place of human staff)
Nano Banana Pro prompted by THE DECODER
OpenAI deploys GPT-6 Astra and Luna as software agents that operate clinic programs and insurance portals using mouse clicks and keystrokes.
In tests, both models reliably handled medical administrative tasks. GPT-6 Luna cost about 8.5 cents per task.
US companies including CVS Health already use the agents for referrals. The offering currently targets the US market and its data privacy requirements.
OpenAI wants GPT-6 Astra and GPT-6 Luna to click and type their way through clinic software and insurance portals as digital clerks. GPT-6 Luna handles administrative tasks for about 8.5 cents per task, according to the company.
In a recent product announcement, OpenAI targets the lack of APIs in healthcare. The computer-use agents integrated into the Agents API in late September operate existing user interfaces the way a human would. In this case, they transfer data between electronic health records and external web portals.
According to OpenAI, when a patient reports back pain, the agent is meant to search relevant prior findings and draft both an MRI order and a reply to the patient. It should also prepare a prior authorization request on the insurer's portal. Clinicians and the responsible administrative staff review these drafts before they are released. The features are available through the API as well as in ChatGPT for Healthcare and ChatGPT for Clinicians.
GPT-6 Luna completes admin tasks for 8.5 cents per task
In an evaluation described by OpenAI, GPT-6 Astra worked through tasks from electronic health record inboxes, among other things preparing an ibuprofen dosage for a child with a broken arm. Physician reviewers rated 99 percent of the 325 tasks evaluated as clinically safe. In 93 percent of 646 reviews, the reviewers called the prep work very or extremely helpful.
Evaluation of GPT-6 Astra’s inbox tasks by physicians using maximum reasoning effort. | Image: OpenAI
For administrative workflows such as prior authorizations, appeals, or medical equipment orders, OpenAI points to the independent HealthAdminBench benchmark from Stanford University researchers. GPT-6 Astra scored 94.3 percent of possible points. GPT-6 Luna, built for low API pricing, scored 91.8 percent and, according to OpenAI, cost about 8.5 cents per task, 98 percent less than Astra. Both models ran at maximum reasoning effort on the benchmark.
Major US healthcare companies test and deploy OpenAI's computer-use agents
Several partner organizations are helping advance the technology. At CVS Health, the agents already process referrals and specialist consultations across separate systems. According to the announcement, one agent cleared a months-long backlog of outgoing documents in one weekend, while another handles the system steps for tens of thousands of consultations daily. CVS Health published its software architecture as Agent Kit on GitHub.
At insurer Cigna Healthcare, the models merge patient information from more than 10 separate systems. Abridge is developing the agents for planned use in clinical discharge planning. Ambience has demonstrated additional safety mechanisms in which an agent stops when a patient's record is unclear and asks a physician for input. Startups like Assort also use the feature in Codex to complete referrals in practice management software whose APIs don't support that final click.
Data privacy, health records, and Europe
Because the agents work directly in electronic health records, protecting sensitive health data is a core concern. OpenAI says it does not use organizational data to train its base models. ChatGPT for Healthcare offers role-based access controls and audit logs, according to the company. Organizations can also configure the applications and environments so an agent only accesses the records and actions needed for a given task.
For HIPAA-compliant workflows with protected health information, organizations must sign a Business Associate Agreement (BAA) with OpenAI. They also need to use the covered services and apply the required security and retention settings. The healthcare facilities themselves are responsible for proper use, but OpenAI, as a contracting party, also has its own obligations under HIPAA. Verified medical professionals in the US who use ChatGPT for Clinicians, a free individual plan, can also sign a BAA, though it applies only to the features the agreement covers.
In the CVS Health deployment, data stays within infrastructure that CVS controls, according to OpenAI. CVS reviews the agents' work steps through screenshots and action logs. The announcement does not say where those screenshots are processed or how long they are stored. It also does not address the European General Data Protection Regulation (GDPR). So far, the offering targets the US market.
OpenAI frames the technology primarily as a tool against administrative friction in a highly fragmented healthcare system. Whether it sees broad adoption likely will not depend solely on the accuracy rates and costs cited in the company's announcement. Also decisive is how reliably the agents connect to existing systems and whether AI agents can be dependably monitored in practice. Another open question is whether a human is required to decide on clinically or legally relevant steps.
AI News Without the Hype – Curated by Humans
原文为英文,正文由自动抓取并保留原文。
原文发布于 The Decoder(AI agents are starting to fill out medical forms and file insurance requests in place of human staff),标题、摘要与正文由机器翻译,术语按行业惯例统一,未经人工逐句校订。