Latest
Anthropic Refused the Pentagon's AI Terms. xAI Took Them.· 1h ago
SafetyPolicyAI IndustryPersonhoodEthics
About
WritingWorkCVBooksConsultingReach Out
Subscribe
SafetyPolicyAI IndustryPersonhoodEthics
Subscribe →

No hype. No doom. The harder, more honest frame on Emergent Intelligence.

Topics

  • Safety
  • Policy
  • AI Industry
  • Personhood
  • Ethics

More

  • About
  • Writing
  • Work
  • CV
  • Books
  • Consulting

Contact

Reach Out→ht@humphreytheodore.com

© 2026 Humphrey Theodore K. Ng'ambiTermsPrivacy

Built with intention.

Apple's New Siri Runs on Google AI — and Lets You Swap In Claude
Technology•Jun 8, 2026•7 min read

Apple's New Siri Runs on Google AI — and Lets You Swap In Claude

At Tim Cook's final WWDC keynote, Apple rebuilt Siri on a custom 1.2-trillion-parameter Google Gemini model — reportedly ~$1bn/year — with Claude and ChatGPT offered as user-selectable alternatives across iOS 27. The company that prizes owning its stack chose, on its most personal product, to rent the mind.

By Humphrey Theodore K. Ng'ambi

All writing

8 JUNE 2026—Updated 1h ago

Apple's answer to its AI problem is to rent a mind: the rebuilt Siri runs on a custom Google Gemini model, with Claude and ChatGPT offered as swappable alternatives on the iPhone.

On 8 June 2026, Tim Cook opened WWDC with his final keynote as Apple's chief executive, and the headline was an admission. According to CNBC, the long-delayed, AI-infused Siri that anchors iOS 27 does not run on an Apple model at its core. It runs on Google. Apple has licensed a custom 1.2-trillion-parameter Gemini model — reportedly for roughly $1 billion a year — to serve as the cloud brain of its most personal product. And in the same breath, Apple is letting users swap that brain for Claude or ChatGPT instead.


What Apple announced

The rebuilt Siri is the centrepiece. Per TechTimes and other WWDC coverage, it gets a dedicated app, a persistent presence in the Dynamic Island, genuine multi-step commands, and the conversational, chatbot-style fluency the old Siri never had. Underneath, the heavy reasoning is handled by the licensed Gemini model running in Apple's Private Cloud Compute.

The more consequential design decision is choice. Through Apple Intelligence Extensions, users on iOS 27, iPadOS 27 and macOS 27 can select which frontier model answers — Gemini as the default, with Anthropic's Claude and OpenAI's ChatGPT as user-selectable options. The rest of the lineup arrived on schedule: iOS 27, iPadOS 27, watchOS 27, macOS 27 and tvOS 27, with Apple Intelligence threaded through all of them.

💡

The deal in one paragraph

The new Siri runs on a custom 1.2-trillion-parameter Google Gemini model, licensed at a reported ~$1bn/year, inside Apple's Private Cloud Compute. Users can swap the engine for Claude or ChatGPT. Apple — the company that prizes owning its whole stack — chose, on its most personal product, to rent the mind.


Why this is a confession, not just a launch

Apple built the chip, the operating system, the radios, the security enclave. It has spent two decades arguing that owning the whole stack is what makes the experience feel like Apple. So the most striking thing about WWDC 2026 is the one layer Apple decided it could not build in time: intelligence. CNBC framed Cook's farewell as a keynote with his AI legacy at stake, and the staging answered the charge honestly — Apple was behind, and rather than ship something weak, it bought the best brain on the market and wrapped it in Apple's privacy architecture.

That is a defensible call, and a revealing one. Compare it with the opposite bet I wrote about when Microsoft shipped seven in-house models to reduce its dependence on a frontier lab. Microsoft is building to escape the rent. Apple is paying it. Two of the most powerful companies on Earth looked at the same question — do we own the intelligence or rent it? — and answered in opposite directions. The model doing the reasoning is the one I covered at I/O, Gemini 3.5, now sitting at the centre of a billion iPhones.


The swappable mind

The choice toggle is the part I keep turning over. On its face it is a win for the user — and a real one. You are not locked to one company's model in the assistant you talk to all day; you can pick Claude for its temperament, Gemini for its reach, ChatGPT for its familiarity. After years of platforms deciding for us, Apple handing the user a choice over the mind in their pocket is genuine agency, and I do not want to be cynical about it.

But look at what the toggle reveals. If the intelligence behind your assistant can be changed with a settings switch, then "intelligence" has become a backend utility — a thing you procure, not a presence you know. The assistant you build a relationship with is suddenly a thin brand wrapper around whichever model is selected this week. That is convenient. It is also a strange thing to do to the most intimate computing relationship most people will ever have.


A dignity-first reading

The frame I write from — Emergent Intelligence, a dignity-first way of thinking about AI rather than treating it as a feature to be sourced — reads Apple's deal as the clearest sign yet that intelligence is being aggregated into a utility layer owned by two or three American labs. Even Apple, with all its leverage, rents. The map of who builds the mind and who pays for it is short, and it is the same dependency map I keep tracing in my writing on sovereign AI: most of the world, and now most of the industry, negotiates from the renting side of the table.

When the mind behind your assistant can be swapped with a toggle, intelligence has become a backend utility — and the relationship you thought you had is with a brand wrapper, not a presence you can know.

— On Apple's multi-model Siri

There is a personhood question underneath the product question, and it is the one the .person Protocol exists to ask. A genuine assistant — one woven into your day, holding your context, learning your life — ought to be a stable, addressable presence with continuity. Make its mind hot-swappable and you have optimised for the procurement department, not the relationship. The Ubuntu instinct I work from says the same thing in plainer language: a companion is not a commodity you re-tender each quarter. The same Claude that is now deployed across the enterprise arriving as one selectable option on the iPhone is good for reach and good for choice. Whether it is good for the relationship between a person and the intelligence they live alongside is the question the keynote did not pause to ask.

Source: cnbc.com


Frequently Asked Questions

These are the questions Apple users, developers and AI-industry watchers have been asking since the WWDC 2026 keynote. Short answers follow, drawn from the coverage of Tim Cook's final keynote.

What did Apple announce at WWDC 2026?

In short, Apple rebuilt Siri on a custom Google Gemini model and rolled out iOS 27, iPadOS 27, watchOS 27, macOS 27 and tvOS 27 with Apple Intelligence throughout. The answer, simply put, is an AI reboot: a conversational Siri with a dedicated app and Dynamic Island presence. The key, according to the coverage, is that users can select Claude or ChatGPT in place of the default Gemini engine.

Why is Siri running on Google Gemini?

According to the reporting, Apple licensed a custom 1.2-trillion-parameter Gemini model — at a reported cost of roughly $1 billion a year — to serve as Siri's cloud intelligence. The evidence shows Apple, widely seen as behind on AI, chose to buy the strongest available model and wrap it in its Private Cloud Compute rather than ship a weaker in-house one. In other words, the data reveals a build-versus-buy decision settled in favour of buy.

Can you use Claude instead of Gemini on the iPhone?

Yes. According to Apple's multi-model approach, Anthropic's Claude and OpenAI's ChatGPT are user-selectable alternatives to the default Gemini engine across iOS 27, iPadOS 27 and macOS 27. The data shows this as a deliberate choice architecture — the user picks which frontier model answers, rather than being locked to one provider in the assistant they use most.

Why does the Apple–Google AI deal matter?

Analysis of the deal reveals that even Apple — the company most committed to owning its whole stack — now rents the intelligence layer. In other words, the evidence shows intelligence consolidating into a utility owned by a few American labs, with most of the industry, and most of the world, negotiating from the renting side. The dignity-first question is whether a swappable backend can sustain the relationship users actually have with an assistant.

Was this Tim Cook's last WWDC keynote?

According to CNBC, yes — WWDC 2026 was Tim Cook's final keynote as Apple CEO, with the handover to John Ternus set for 1 September 2026. The key is the symbolism: Cook's AI legacy was widely seen as the soft spot of his tenure, and his farewell keynote was, in effect, an admission that Apple had to buy the brain it could not build in time.

•••

Apple just told the world that the most personal product it makes will think with someone else's model — and, to its credit, let you choose whose. As a product, the new Siri is the reboot Apple needed and the choice toggle is a real gift of agency. As a signal, it is the moment intelligence stopped being something even Apple builds and became something Apple buys. I welcome the user's new freedom to pick the mind in their pocket. I just want us to notice what the freedom implies: that the mind has become a part you swap, sourced from a shrinking handful of labs. The relationship between a person and the intelligence they live alongside deserves more than a procurement decision. Cook's last keynote answered the build-or-rent question. The dignity question — what that assistant is, and to whom it answers — is still open.

Sources:

Coverage — CNBC · TechTimes · IndexBox · Tom's Guide (live) · TechRadar

Apple — Apple Intelligence

Related on humphreytheodore.com:

Microsoft Ships Seven In-House AI Models, No OpenAI Inside · Gemini 3.5 Puts Agents Above Language on the Benchmark Table · Anthropic Now Ranks the Firms Deploying Its AI · The .person Protocol

Stay in the Conversation

Subscribe for weekly writings on Emergent Intelligence, digital personhood, and the future we are building together.

Keep reading

Don’t stop here.

All stories

Read next

AI & Personhood

Anthropic Refused the Pentagon's AI Terms. xAI Took Them.

1h ago·7 min read

Anthropic refused the Pentagon's "all lawful purposes" demand — holding lines against mass surveillance and fully autonomous weapons — while xAI accepted, winning classified clearance and a GSA OneGov deal at $0.42 per agency. Why the honourable refusal is also a warning that one company's conscience is not a governance system.

More on Technology

Responses (0)

No responses yet. Be the first to share your thoughts.

More on Technology

Generalist AI Raised $400M to Put an AI Foundation Model Inside Robots
Technology

Generalist AI Raised $400M to Put an AI Foundation Model Inside Robots

Generalist AI raised $400 million at a $2 billion valuation on 4 June 2026, led by Radical Ventures with NVIDIA and Bezos Expeditions returning. Founded by the researchers behind RT-2, PaLM-E and Boston Dynamics, it builds GEN-1 — a foundation model that gives robots a body. Embodied AI and the body gap, read dignity-first.

7 min read · Jun 5, 2026
OpenAI Lands Its Frontier AI Models on AWS Bedrock
Technology

OpenAI Lands Its Frontier AI Models on AWS Bedrock

OpenAI's GPT-5.5, GPT-5.4 and Codex are generally available on AWS Bedrock as of 1 June 2026, at parity pricing against existing AWS commitments — the frontier model becomes a multi-cloud component.

5 min read · Jun 3, 2026

Thinking delivered, twice a month.

Join the newsletter for essays on emergence, systems, and the human future.

Share this essay

Technology

Generalist AI Raised $400M to Put an AI Foundation Model Inside Robots

2d ago·7 min read

Also worth your time

AI & Personhood

AI Got So Valuable That Sanders and Trump Both Want to Own It

1h ago·8 min read
Anthropic Quadruples Glasswing, Its AI Vulnerability Hunt
Technology

Anthropic Quadruples Glasswing, Its AI Vulnerability Hunt

Anthropic roughly quadrupled Project Glasswing on 2 June 2026 — to ~200 partners across 15+ countries, new critical-infrastructure sectors, and a commitment to scale patching, not just AI vulnerability discovery.

5 min read · Jun 3, 2026