Apple's Siri AI Can Be Swapped Out for Claude, ChatGPT, Code Shows
代码侦探在 iOS 27 和 macOS Golden Gate 的私有框架中发现了证据，表明 Apple 已将 Siri 架构设计为在很深的层面上支持第三方 AI 模型。该发现由用户 pdfu 披露，揭示了允许外部 AI 服务直接接入 Siri 体验的机制。
其中一个关键机制是 Model Delegation，允许像 Claude 这样的第三方模型作为 Siri 的扩展运行。演示中，用户可以通过系统搜索栏的上下文菜单选择外部模型；当发出请求时，所选模型会解析用户意图，并在必要时将任务交给 Siri 去在 Apple 应用中执行诸如设置提醒或创建文件等操作，这些是 Siri 单独可能无法完成的。
更重要的是，在 Model Manager Services 中发现的一项协议显示，第三方模型（例如 GPT-5.6）有可能完全替代 Apple 服务器端的 Siri 模型。在这种模式下，外部模型会接收 Apple 原生的 Siri planner 提示和工具定义，从而能够执行系统操作、访问相关个人数据，并生成通过熟悉的 Siri 界面和语音呈现给用户的回复。
这种更高程度的互操作性可能受到了 European Union 的 Digital Markets Act 的推动，该法要求 Apple 向第三方提供对其软硬件功能的有效访问。尽管 Apple 尚未向公众或第三方开发者开放这些具体的模型委派权限，但相关基础设施的存在表明，Apple 在设计软件时已为未来的 AI 模型灵活性预留了空间。
目前这些能力对普通用户还不是完全开放，发布版本中"Ask"功能主要仅接入了 ChatGPT 。但这些底层协议暗示了一种重大架构转变：Siri 有望变得更加模块化，用户可以选择最适合自己需求的 AI 模型，同时仍享有深度的系统集成。
Code sleuths have uncovered evidence in the latest iOS 27 and macOS Golden Gate private frameworks suggesting that Apple has designed its Siri architecture to support third-party AI models at a remarkably deep level. This discovery, highlighted by a user named pdfu, reveals mechanisms that allow external AI services to integrate directly into the Siri experience.
One of the primary mechanisms identified is Model Delegation, which permits third-party models like Claude to function as a Siri extension. In a practical demonstration, a user can select an external model through a contextual menu in the system's search bar. When a request is made, the chosen AI model interprets the user's intent and, if necessary, hands the task back to Siri to perform actions within Apple apps, such as setting reminders or creating files that Siri might not be able to handle on its own.
An even more significant discovery involves a protocol found within Model Manager Services that enables the complete replacement of Apple's own server-side Siri model with third-party alternatives, such as GPT-5.6. Under this setup, the external model receives Apple's native Siri planner prompts and tool definitions. This allows the third-party model to execute system actions, access relevant personal data, and formulate responses that are then presented to the user through the familiar Siri interface and voice.
This shift toward greater interoperability may be influenced by the European Union's Digital Markets Act, which mandates that Apple provide third parties with effective access to its software and hardware features. While Apple has yet to open these specific model delegation entitlements to the general public or third-party developers, the existence of this infrastructure demonstrates that the company has engineered its software with a focus on future AI model flexibility.
Currently, these capabilities are not fully front-facing for the average user, and the "Ask" functionality is primarily limited to ChatGPT in the release version of the new software. Nevertheless, these underlying protocols suggest a significant architectural change that could reshape how Siri operates, potentially allowing for a more modular approach where users can choose the AI models that best suit their needs while still benefiting from deep system integration.
117 comments • Comments Link
• Siri 未能抢占先机，因为它没有向开发者开放，错失了通过用户创建的 agents 和 automations 推动生态演进的重大机会。
• 语音助理在很大程度上陷入僵局，用户更看重简单、可靠的任务（例如控制家居设备或执行基本指令），而不是那些通常缺乏实用性的复杂 AI 聊天交互。
• 目前的 Siri 架构在很大程度上受到 EU regulations 和 Digital Markets Act 要求的影响，促使 Apple 构建了模块化、可能可替换的 AI framework，从而在保持对用户体验控制的同时应对监管要求。
• Apple 决定构建解耦的 AI framework（模型可以被替换或更新），这是一项战略必要，使其能够在 on-device 、 hybrid 与 cloud-based 模型之间切换，而无需重建整个基础设施。
• Siri 缺乏处理高级、多步骤指令的能力，尤其在上下文保持和复杂的智能家居编排方面，落后于 Google Assistant 或 Home Assistant 等替代品。
• Apple 似乎有意在 EU 限制新 AI 功能的推广，作为一种谈判策略，以抵制授予第三方 AI 提供方对系统级数据和硬件无限制访问的要求。
• 市场对本地 AI "broker" 或协调器存在明确需求，这类产品可以将用户选择的 LLMs 连接到设备硬件，并将敏感数据保留在私有网络边界内。
• 虽然 Apple 目前在其 AI 功能上依赖 white-labeled models，但其长期战略更侧重于掌控用户界面和数据隐私，利用其庞大的分发渠道保持竞争力，而无需在初期的 AI 模型竞赛中获胜。
• 许多用户对语音助理仍持怀疑态度，更倾向于一个能可靠处理基本任务的"哑"系统，而不是一个可能侵犯隐私、泄露用户数据或默认使用基于广告的第三方服务的复杂 AI 。
• 开发者认为 Apple 为 AI 设计的模块化 API 封装只是大型平台的工程常规做法，能确保系统在面对硬件变化和不断演变的法律环境时更具前瞻性。
这场讨论反映出人们对复杂、具代理性的 AI 助手的渴望，与现有语音控制界面现实且常令人失望的表现之间的张力。许多人将 Apple 最近的架构选择视为受 EU 监管压力驱动的防御性举措，但也有人认为这种模块化设计是一种战略性尝试（尽管为时已晚），旨在构建一个最终能够集成第三方模型的灵活生态。归根结底，达成的共识是"AI as a chatbot"并不是令人振奋的目标，真正的价值在于与本地硬件以及用户定义的、注重隐私的工作流进行深度整合。 • Siri failed to gain an early advantage because it was not opened to developers, missing a massive opportunity for the ecosystem to evolve through user-created agents and automations.
• Voice assistants have largely struggled because users prioritize simple, reliable tasks—like controlling home hardware or basic commands—over complex AI chatbot interactions, which often fail to provide actual utility.
• The current Siri architecture is heavily influenced by the need to comply with EU regulations and the Digital Markets Act, leading Apple to build modular, potentially swappable AI frameworks that keep the company in control of the user experience.
• Apple's decision to build a decoupled AI framework—where models can be swapped or updated—is a strategic necessity that allows them to pivot between on-device, hybrid, and cloud-based models without rebuilding their entire infrastructure.
• The lack of advanced, multi-step command handling in Siri has left it trailing behind alternatives like Google Assistant or Home Assistant, particularly regarding context retention and complex smart home orchestration.
• Apple appears to be intentionally limiting the rollout of new AI features in the EU as a negotiation tactic, resisting requirements to grant third-party AI providers unfettered access to system-level data and hardware.
• There is a clear market demand for a local AI "broker" or orchestrator that could connect user-chosen LLMs to device hardware, keeping sensitive data within private network boundaries.
• While Apple is currently relying on white-labeled models for its AI features, its long-term strategy focuses on maintaining control over the user interface and data privacy, effectively using its massive distribution base to stay competitive without needing to win the initial AI model race.
• Many users remain skeptical of voice assistants and would prefer a "dumb" system that handles basic tasks reliably, rather than a sophisticated, privacy-invasive AI that risks leaking user data or defaulting to third-party ad-based services.
• Developers argue that Apple's design of modular API wrappers for AI is simply standard engineering practice for large platforms, ensuring that the system remains future-proof against both hardware shifts and evolving legal environments.
The conversation reflects a tension between the desire for a sophisticated, agentic AI assistant and the practical, often disappointing reality of existing voice-control interfaces. While many see Apple's recent architectural choices as a defensive maneuver driven by EU regulatory pressure, others view the modular design as a strategic, albeit late, attempt to build a flexible ecosystem that can eventually integrate third-party models. Ultimately, there is a clear consensus that "AI as a chatbot" is an uninspiring endpoint, and the true value lies in deep integration with local hardware and user-defined, privacy-conscious workflows.