Apple Intelligence: On-Device AI, Siri and ChatGPT Handoff

Apple's own system, built into the operating system

Apple announced Apple Intelligence on June 10, 2024, describing it as a personal intelligence system built into iOS 18, iPadOS 18 and macOS Sequoia. It combined generative models with a user's on-device personal context to draft and rewrite text, summarize notifications, generate images, and let apps take actions across each other. [1]

Apple said most requests would run on-device using Apple silicon, with the remainder handled by Private Cloud Compute, a set of Apple-operated servers it described as never retaining or exposing user data and open to inspection by independent researchers checking that only the published software runs there. [1]

One deliberate exception, for the questions Siri can't answer

Inside a system Apple built and controlled end to end, it carved out one named exception. Siri and the Compose feature in Writing Tools could hand a query to ChatGPT, running OpenAI's GPT-4o, but only after asking the user's permission first. [1]

Apple said the handoff would work without a ChatGPT account, that a paid subscriber could link their own account for higher limits, and that OpenAI would not log the requests and would mask the user's IP address before the query arrived. [1]

Atlas interpretation: Apple's usual pitch is a stack it owns from chip to app. Routing the moment its own assistant runs out of answers to a rival's model, permission prompt or not, is Apple naming a capability gap in public rather than papering over it with a weaker model of its own. [1]

Announced in June, shipped in pieces later

None of what Apple showed at the keynote was available that day. The announcement described Apple Intelligence as arriving through a free software update later in 2024, in beta, and limited at first to American English. [1]

The first Apple Intelligence features reached devices with iOS 18.1 in late October 2024. The ChatGPT integration and other capabilities Apple had demonstrated in June followed in subsequent point releases rather than at that initial launch. [2]

The Siri feature that didn't ship, and the Gemini deal that followed

Atlas interpretation: The capability Apple emphasized most in June 2024, a Siri that could reach into a user's own apps and data to act on personal context, was not part of this release. Apple described it only as coming in a future update, and that promise slipped for years rather than months. [2]

In January 2026, Apple agreed to base its next Apple Foundation Models on Google's Gemini models rather than its own. That deal left the opt-in ChatGPT handoff for Siri queries in place alongside it, rather than replacing it. [2]