Siri vs. ChatGPT vs. Gemini: Which AI Is Safest With Your Data?

This week, three of the biggest AI platforms each made a move that tells you exactly how they think about your data.

Apple announced that its upcoming Siri overhaul — a standalone app coming out of beta this fall — will auto-delete your conversation history by default. You get the same controls that exist in iMessage: delete after 30 days, a year, or keep everything. Apple will also run its Siri intelligence on Private Cloud Compute servers, even when using a third-party model like Google's Gemini underneath.

Meanwhile, Google confirmed that Gemini Intelligence is now the ambient layer across Android, ChromeOS, Wear OS, and Android Auto — reading your screen context, your notifications, and your messages to provide proactive assistance across three billion devices.

And OpenAI launched ChatGPT Personal Finance, which connects directly to your bank account through Plaid to analyze your spending. What OpenAI does with that financial data for model training is not clearly addressed in its documentation.

These are not random product announcements. They are three very different answers to the same question: how should an AI handle what you share with it?

Apple: the ephemeral approach

Apple's bet is that you will trust an AI more if it does not permanently remember everything you have said. The auto-delete feature in the new Siri app mirrors the same controls iMessage already offers, so if you are already a heavy Messages user, you know exactly how it works.

The more technically significant piece is Private Cloud Compute. When Siri needs to use a more capable model — in some cases, Google Gemini — Apple routes that request through its own servers rather than sending your query directly to Google. Apple has published its cryptographic attestation process for Private Cloud Compute, which lets security researchers verify that queries are not logged or retained. That is a meaningful transparency step that neither Google nor OpenAI has matched at this level of detail.

The tradeoff is real: if the AI does not remember your conversations, it cannot build a model of your preferences, habits, or history over time. You start fresh, or close to it, on every interaction. That is a deliberate design choice, not a limitation they overlooked.

Apple's new Siri app is expected to debut at WWDC 2026 (June 8) and launch in beta on iOS 27. 9to5Mac has a detailed breakdown of what is confirmed so far.

Google: the ambient approach

When Google says Gemini is now part of your Android operating system, they mean it literally. Gemini Intelligence can see your screen, read your notifications, and understand the context of whatever app you are currently using. The feature launches this summer across Android phones, tablets, Wear OS watches, car dashboards through Android Auto, and mixed reality headsets via Android XR.

As we covered in our breakdown of the Android Show announcement, this is distribution by default rather than by user adoption. You do not choose to use it. It is already there.

Google's privacy controls for Gemini include the ability to review and delete activity, pause data collection, and opt out of using your conversations to improve Google's models. But the data collected while Gemini is running as an OS-level service is broader than what you would share by opening a standalone app. The surface area of what gets processed is substantially larger, and those controls require you to know where to look.

OpenAI: the connected approach

OpenAI's ChatGPT Personal Finance is the most direct example of an AI accessing genuinely sensitive personal data. The Plaid integration works by generating a tokenized access grant rather than sharing your actual banking credentials. ChatGPT receives four categories: account balances, transaction history, merchant names, and transaction amounts.

The data connection itself is more controlled than older screen-scraping approaches. But as our analysis of the feature found, the training-data question stays open. OpenAI's documentation says data may be used "to improve services." Whether financial query data specifically influences model training is not addressed.

ChatGPT does have a privacy toggle — "Improve the model for everyone" in Settings — that can be turned off. The scope of that setting for new connected-data features like Personal Finance is not clearly defined.

Anthropic: the stated approach

Claude does not have the ecosystem play that Apple, Google, and OpenAI are building. But Anthropic publishes a detailed privacy policy stating that Claude conversations are not used to train models by default, and enterprise customers get additional data isolation guarantees. If you are evaluating AI assistants specifically on privacy commitments, Claude's data handling approach is a useful baseline.

What this means for you

If you are deciding which AI assistant to use — or which to give deeper access to your data — here is a practical framework.

For everyday tasks, all three options are roughly comparable in risk. The privacy differences matter most when you are discussing something sensitive: medical information, financial decisions, personal relationships, or legal questions.

For sensitive conversations, Apple's auto-delete is the most defensible option available today. The data does not persist by default. Google and OpenAI both retain data unless you actively manage your settings. Most people never do.

For professional or financial tasks, the relevant question is not just which AI you are using. It is what you are connecting to it. A chatbot with access to your bank account, your calendar, and your email is not the same product as a chatbot you type questions into. The risk profile scales directly with the connections you authorize.

Apple is betting that privacy-first design is a competitive advantage worth building toward. Whether that translates into market share will depend on whether enough users treat privacy as a real selection criterion rather than a stated one.

Right now, most of the data suggests they do not. Apple is betting that changes.

---

Every week, About.chat Weekly rounds up the most important AI developments — no hype, no fluff. Subscribe free at aboutchat.beehiiv.com. It takes two minutes and you can unsubscribe anytime.