Apple spent years positioning itself as the privacy company. iCloud data stays on Apple’s servers. Siri processing happens on-device or, when it has to go to the cloud, through Private Cloud Compute — Apple’s isolated, independently auditable environment.
Then came iOS 27, announced at WWDC 2026, and a new privacy question arrived with it: what happens when you set Gemini, Claude, or ChatGPT as your preferred Siri AI model, and Siri passes your personal context to them?
What iOS 27 Actually Changed
iOS 27 introduces an AI Extensions framework that lets users choose a third-party AI model to power certain Siri requests. According to Bloomberg’s reporting ahead of WWDC, the implementation is expected to be opt-in: you set your preferred provider in Settings, and Siri routes complex or generative requests to that model rather than handling them internally.
Beta testing showed the feature is available in the iOS 27 developer build, though activation depends on region and provider agreements still finalising as of early autumn 2026.
The practical effect is significant. Siri, in its current implementation, can access a wide range of personal data: your calendar events, your messages, your notes, your reminders, your health summaries, your HomeKit routines. That data informs the “context” Siri passes with a request.
When the AI provider is Apple, that context stays within Apple’s infrastructure and Apple’s privacy commitments. When you select a third-party provider, the context leaves Apple’s servers and travels to Google’s, Anthropic’s, or OpenAI’s infrastructure — each with its own privacy policy, retention terms, and legal exposure.
What “Personal Context” Actually Includes
Apple’s Siri personal context is richer than many users realise. Depending on the apps installed and permissions granted, it can include:
- Calendar events: every appointment, meeting, and personal event in your calendar
- Messages: recent conversations and their participants
- Notes: content from the Notes app
- Health summaries: data from Health app if Siri has access
- HomeKit state: your smart home routines and schedules
- App activity: recent app usage patterns Siri learns from
Not all of this necessarily travels with every request. Apple says context is passed only as relevant to the specific query. But the determination of “relevance” is made algorithmically, and users do not have a line-by-line view of what accompanies each handoff.
Apple’s Stated Privacy Protections for Third-Party Handoffs
Apple has not remained silent on this. The company says that when requests are routed to third-party AI providers, the privacy protections of the Private Cloud Compute model still apply — requests are processed inside isolated confidential virtual machines, and the provider cannot retain the data beyond the request window.
Apple also says these claims are externally verifiable, meaning security researchers can audit the infrastructure in ways they cannot with most cloud services.
There are important caveats, however:
Audits verify infrastructure, not policy. An audit confirming that a server environment is isolated does not confirm what the AI provider does with the inferred outputs of processing your request. The output of analysing your personal context — the model’s understanding of your question and its answer — may still inform the model’s training or the provider’s internal analytics, depending on the provider’s terms.
Terms vary by provider. Apple controls its own Private Cloud Compute terms. Google, Anthropic, and OpenAI each have their own data use policies, and those policies differ in how they handle data received through integrations. Reading the specific provider’s API or integration terms — not just their consumer product terms — is where the relevant language lives.
The feature is new. WWDC 2026 was the announcement; iOS 27 is in early public release. Privacy researchers and regulators have not yet fully analysed the implementation. The EU has raised questions about the Apple-OpenAI relationship going back to iOS 18; iOS 27’s broader third-party model is a newer subject of scrutiny.
What to Do Before Enabling a Third-Party AI Provider
If you are considering enabling this feature — or already have it set up — a few checks are worth doing first.
Read the provider’s integration terms, not their marketing page. The consumer privacy policy on the provider’s website describes how they handle data you enter directly into their product. The relevant question is how they handle data received as API context from a third party. These can differ.
Set the feature to specific request types only. iOS 27 lets you configure which request categories route to the external provider. Routing writing assistance tasks to a third-party model carries different implications than routing health queries. Segment the use accordingly.
Check your Siri data access settings. In iOS Settings → Privacy & Security → Siri & Search, you can review which apps Siri has access to. Removing access for apps whose data you do not want in the context that travels externally reduces the potential scope of any handoff.
Understand that context syncs through iCloud. Siri’s personal context is synchronised across your Apple devices via iCloud. This means context built on your iPhone is also available on your iPad and Mac when a Siri query is handled. The broader the device footprint, the broader the context.
The Fundamental Question: Who Do You Trust With Your Context?
Apple’s Private Cloud Compute represented a genuine architectural privacy commitment. The decision to route requests to a third-party AI model is a departure from that architecture — even if the implementation preserves some of its properties.
This is not an argument against the feature. Third-party AI providers may offer genuinely better responses for certain tasks. The argument is for making the choice consciously rather than defaulting into it because the option was surfaced in settings and it seemed simpler to say yes.
Every time you ask Siri to do something that requires understanding your personal life — summarising an email thread, managing a reminder based on your past behaviour, suggesting an action based on your calendar — that data may now travel to servers owned and operated by a company whose interests are not identical to Apple’s.
The Broader Pattern in Apple’s Privacy Strategy
Apple has been navigating a genuine tension for several years. On-device AI provides strong privacy guarantees but limited capability. Cloud AI from sophisticated providers offers more powerful responses but weaker guarantees.
Private Cloud Compute was Apple’s attempt to split the difference within its own infrastructure. iOS 27’s third-party AI extensions extend that to providers Apple does not control, which is a meaningful change in the privacy model regardless of the safeguards Apple has built around the handoff.
Users who care about this distinction should watch how the feature evolves. The initial implementation is cautious — opt-in, auditable, limited in scope. The question is whether those constraints hold as the feature matures and as commercial incentives around AI partnerships grow.
Keeping Personal Files Outside the AI Context Loop
One response to this landscape is to keep your most private personal files in storage that exists outside the AI ecosystem entirely — not connected to Siri, not accessible as context for any AI query, not subject to handoff decisions.
daftei stores photos, notes, documents, and voice memos with AES-256 encryption at rest and TLS 1.3 in transit. It operates on iOS, Android, and the web at /app. It is GDPR and CCPA compliant, never sells user data, never shows ads, and never uses stored content to train AI models.
Files stored in daftei are not visible to Siri, cannot be included in Siri’s personal context, and do not travel with any AI request — because they are not connected to the AI layer at all.
That is a trade-off in one direction: you do not get AI-powered search or automatic memory features for those files. You get a private archive that stays private regardless of what AI features you choose to enable elsewhere on your device.
Both models have a place. Knowing which files belong in which kind of storage is the practical skill iOS 27 makes newly important.