Can my iPhone app send user data to a third-party AI model?

Apple's App Store rule on sharing personal data with third-party AI, what clear disclosure and explicit permission look like in practice, and a checklist an agent can run against your app.

Short answer

Yes, but on iOS you must first tell users clearly where their data is going and get their explicit permission. On November 13, 2025, Apple added this sentence to Guideline 5.1.2(i) of its App Review Guidelines: "You must clearly disclose where personal data will be shared with third parties, including with third-party AI, and obtain explicit permission before doing so." The rule covers sharing personal data with any third party, and it names AI providers expressly. It is an App Store requirement, not a law, so it applies in addition to the privacy laws discussed in Can I send user data to an LLM provider?

What the rule reaches

• Third-party AI. The rule applies when personal data goes to a model provider outside your company. AI processing done entirely on the device, with no personal data sent to another organization, generally isn't third-party sharing simply because it uses AI. But using an Apple-provided AI framework doesn't by itself mean the processing stays on the device, and the guideline doesn't exempt Apple's models across the board. What matters is where the data actually goes.
• Personal data. Read it broadly: messages, photos, voice, health and fitness data, contacts, location, and anything the user typed that identifies them or someone else.
• Explicit permission. An affirmative choice by the user, such as tapping a clearly labeled button, made before the first transfer. A sentence in the privacy policy or terms does not count, and neither does a pre-checked box.

What else has to line up

• New uses need new consent. Guideline 5.1.2(ii) prohibits using data you have already collected for a new purpose without further consent, unless the law otherwise allows it.
• Health data. Guideline 5.1.3 restricts data from health, fitness, and medical-research contexts, including HealthKit, Clinical Health Records, and Motion and Fitness data. That data generally can't be used or disclosed for advertising, marketing, or other use-based data mining. Apple's HealthKit rules also allow sharing with third parties only with the user's express permission and, outside medical research, only when the recipient also provides a health or fitness service to the user. General consent doesn't override these limits, so health data in prompts needs its own review.
• App privacy details. The data types and uses you declare in App Store Connect must match what the app actually sends to the AI provider.
• Your privacy policy. It should state the categories of data sent to AI providers, the purpose, and either the providers' names or the type of provider.
• Your contract with the provider. Check the provider's retention and training terms, and any zero-data-retention option, so that what you tell users is true.
• Google Play. Its User Data policy requires prominent in-app disclosure and affirmative consent when collecting, using, or sharing personal or sensitive data in ways users wouldn't reasonably expect. The disclosure must appear in the app's normal flow and say what data is involved and how it's used or shared; a privacy policy alone isn't enough. Build one flow that satisfies both stores.

A design that works

• A consent screen before the first AI use, not at app install. Say what data is sent, to which provider, and why, in a sentence or two.
• A real choice. Let users decline and keep using the parts of the app that don't need AI.
• Nothing sent before consent. That includes "warm-up" requests, analytics on prompt content, and prefetching.
• A record of consent. Store when the user agreed and which version of the disclosure they saw.
• A way to withdraw consent in Settings that stops future transfers.
• Reassess when things change. When you change AI providers, send new categories of data, or add a purpose such as training, check whether your existing disclosure and permission still cover what you're doing. New purposes need further consent. If the existing permission doesn't cover a new provider or data flow, get renewed disclosure and consent before the change ships.

Checklist an agent can run against the codebase

• Find every network call to a model provider and list the data fields each one sends.
• For each call, find the consent check that gates it, or note that none exists.
• Check that no personal data is sent before consent, including in background tasks and retries.
• Check which AI processing happens on the device and which sends data off it, including through Apple frameworks.
• Find the consent screen text and compare it with the actual data sent and the provider used.
• Check that declining consent leaves the rest of the app usable.
• Check that consent is recorded with a timestamp and disclosure version, and can be withdrawn in Settings.
• Check whether any HealthKit or other health data can reach a prompt.
• Compare the App Store Connect privacy details and the privacy policy with the data actually sent.

Sources

• Apple App Review Guidelines, sections 5.1.2 and 5.1.3, and Apple's November 13, 2025 announcement of the updated guidelines (developer.apple.com)
• Google Play User Data policy (Play Console Help)

If this applies to you

An AI feature review covers the provider terms, disclosures, consent flow, and app store requirements together. Adding AI to an existing app is also a good fit for a pre-build feature check.