Skip to content
← Writing
Writing · October 2026 · 2 min read

The most private AI never leaves the phone

Google DeepMind and Singapore's GovTech built a scam detector that runs on the phone itself. When sensitive work should stay on the device, and when it should not.

Abhijit Chatterjee Chattopadhyay

The Logic Atelier logo above the words: The most private AI never leaves the phone.

On 14 October, Google DeepMind and the Government Technology Agency of Singapore described a scam detection model that runs on the phone itself. Their post says that because it runs entirely on the device, neither the government nor Google will be able to read users' conversations through it.

That is from Google DeepMind's post on the project (opens in a new tab). The model is not rolled out yet. The post says the rollout will be phased. But the design answers a common question: does sensitive data have to go to the cloud for AI to work on it?

Why they kept it on the phone

Scam detection means reading private messages. Sending those to a server to be checked would create the very exposure the tool is meant to prevent. So they took the model to the data instead.

They started from Gemma, Google's lightweight model, and trained it further on local scam conversations, because "Generic models often struggle with local deception techniques." A small model that knows the local tricks can do this job where a large general one might not.

To keep the battery alive, a light first check screens incoming messages and only passes likely threats to the fuller model. That is the same broad-first, careful-second design I described in The small model that beat the LLM.

Three questions before you choose where AI runs

On-device AI is not right for everything. These three questions tell you when it is.

  1. Would sending the data out create the risk you are trying to avoid? Private messages, health notes, client files. If the answer is yes, start by asking whether the work can stay where the data already is.
  2. Does it need an answer in the moment? A warning before someone sends money is useless a minute late. On the device, there is no network round trip and no outage on someone else's server.
  3. Is the job narrow? Spotting one kind of threat is narrow. Writing a report is not. Small models that fit on a phone or a laptop do narrow jobs well and broad ones poorly.

Three yeses, and the device may be the better place for the work. If the job is broad, the cloud is still the right answer, with care over what you send.

Do this today

  • List the AI tasks in your business that touch private or client data.
  • For each one, write down what data leaves your systems and where it goes.
  • Mark the narrow ones. Those are the candidates to run locally.

Your next step

I write one practical note like this every day. Sign up just below to get them by email.

Drafted with AI help, edited and approved by me.

New articles by email

I will never sell your details, for any reason, and I never send spam.

How I use your details
  • I, Abhijit Chatterjee Chattopadhyay of The Logic Atelier, hold them, in my own newsletter system hosted on Fly.io. Emails go out through Resend.
  • Your name is for greeting you, your email is for sending new articles, and your phone number, if you give it, is so I can reach you directly about my work.
  • If you do not confirm your email within 7 days, I delete the sign-up. If you unsubscribe, I delete your details 30 days later.
  • If Google Analytics is running on your visit, it is told that a sign-up was made, and nothing you typed.
  • To see or delete what I hold about you at any time, email hello@thelogicatelier.com.
  • The full detail is on the privacy page.
NextYour AI costs grow with use. Does your price?
← cd ../writing

The most private AI never leaves the phone

on-device-ai.mdfront matter
title: The most private AI never leaves the phone
date: 2026-10-30
reading_time: 2 min
author: Abhijit Chatterjee Chattopadhyay
on_device_ai_cover_92dbc1a5_af50bf5683.pngcover
The Logic Atelier logo above the words: The most private AI never leaves the phone.

On 14 October, Google DeepMind and the Government Technology Agency of Singapore described a scam detection model that runs on the phone itself. Their post says that because it runs entirely on the device, neither the government nor Google will be able to read users' conversations through it.

That is from Google DeepMind's post on the project (opens in a new tab). The model is not rolled out yet. The post says the rollout will be phased. But the design answers a common question: does sensitive data have to go to the cloud for AI to work on it?

Why they kept it on the phone

Scam detection means reading private messages. Sending those to a server to be checked would create the very exposure the tool is meant to prevent. So they took the model to the data instead.

They started from Gemma, Google's lightweight model, and trained it further on local scam conversations, because "Generic models often struggle with local deception techniques." A small model that knows the local tricks can do this job where a large general one might not.

To keep the battery alive, a light first check screens incoming messages and only passes likely threats to the fuller model. That is the same broad-first, careful-second design I described in The small model that beat the LLM.

Three questions before you choose where AI runs

On-device AI is not right for everything. These three questions tell you when it is.

  1. Would sending the data out create the risk you are trying to avoid? Private messages, health notes, client files. If the answer is yes, start by asking whether the work can stay where the data already is.
  2. Does it need an answer in the moment? A warning before someone sends money is useless a minute late. On the device, there is no network round trip and no outage on someone else's server.
  3. Is the job narrow? Spotting one kind of threat is narrow. Writing a report is not. Small models that fit on a phone or a laptop do narrow jobs well and broad ones poorly.

Three yeses, and the device may be the better place for the work. If the job is broad, the cloud is still the right answer, with care over what you send.

Do this today

  • List the AI tasks in your business that touch private or client data.
  • For each one, write down what data leaves your systems and where it goes.
  • Mark the narrow ones. Those are the candidates to run locally.

Your next step

I write one practical note like this every day. Sign up just below to get them by email.

Drafted with AI help, edited and approved by me.

subscribeemail

// I will never sell your details, for any reason, and I never send spam.

How I use your details
  • I, Abhijit Chatterjee Chattopadhyay of The Logic Atelier, hold them, in my own newsletter system hosted on Fly.io. Emails go out through Resend.
  • Your name is for greeting you, your email is for sending new articles, and your phone number, if you give it, is so I can reach you directly about my work.
  • If you do not confirm your email within 7 days, I delete the sign-up. If you unsubscribe, I delete your details 30 days later.
  • If Google Analytics is running on your visit, it is told that a sign-up was made, and nothing you typed.
  • To see or delete what I hold about you at any time, email hello@thelogicatelier.com.
  • The full detail is on the privacy page.

vi.Contact

06 · contact

Talk to me.

Talk to me.

Production AI and software architecture. Based in Dubai, working with teams in the Emirates and Ireland.

I work Dubai hours, Monday to Friday, the same week you do. Dubai is three hours ahead of Dublin in summer and four in winter, so my afternoon is your morning.

I will never sell your details, for any reason, and I never send spam.

How I use your enquiry
  • Your enquiry comes to me, Abhijit Chatterjee Chattopadhyay of The Logic Atelier. The form is handled by Formspree, which passes it to my inbox and keeps a copy in my Formspree account.
  • I use it to read your enquiry and reply. If no work follows, I delete it within 12 months, and sooner whenever you ask.
  • If Google Analytics is running on your visit, it is told that an enquiry was sent, and nothing you typed.
  • To see or delete what I hold about you, email hello@thelogicatelier.com.
  • The full detail is on the privacy page.