Secrets in a Flutter app: what you can store, and what you cannot

Start with the uncomfortable part, because everything else follows from it.

Any string compiled into your app is public. Not “hard to find” — public. An APK is a zip file; strings on the extracted binary takes seconds. --dart-define values, constants, obfuscated names, base64-encoded blobs: all of it is recoverable by anyone motivated enough to download your app once.

This is not a Flutter weakness. It is true of every client application on every platform. What it changes is where you draw the line between “the app knows this” and “the app can ask for this”.

The line

Kind of secret	Where it belongs
Third-party API key with billing attached	Server only. The app calls your backend, your backend calls them
Public/publishable keys (Stripe publishable, Firebase config, Maps key)	In the app — they are designed for it, and restricted server-side
User session token	Device secure storage, short-lived, refreshable
Refresh token	Device secure storage, revocable server-side
Encryption key for local data	Derived or generated on-device, stored in Keychain/Keystore
Signing keys, service accounts	Never in the repository, never in the app

The Firebase row surprises people. google-services.json is not a secret — its contents are visible in any app that ships it, and Firebase’s security model is built on server-side Security Rules, not on hiding the config. If your Firestore is protected only by the config file being “hidden”, it is not protected.

The general test: if leaking the value lets an attacker do something as you rather than as themselves, it cannot live on the device.

Using platform secure storage

flutter_secure_storage wraps iOS Keychain and Android’s EncryptedSharedPreferences behind one API:

final class TokenStore {
  TokenStore(this._storage);

  final FlutterSecureStorage _storage;

  static const _accessKey = 'access_token';
  static const _refreshKey = 'refresh_token';

  Future<void> save({
    required String access,
    required String refresh,
  }) async {
    await Future.wait([
      _storage.write(key: _accessKey, value: access),
      _storage.write(key: _refreshKey, value: refresh),
    ]);
  }

  Future<String?> readAccess() => _storage.read(key: _accessKey);
  Future<String?> readRefresh() => _storage.read(key: _refreshKey);

  Future<void> clear() => _storage.deleteAll();
}

Configure the platform options explicitly rather than accepting defaults:

final storage = FlutterSecureStorage(
  aOptions: const AndroidOptions(encryptedSharedPreferences: true),
  iOptions: const IOSOptions(
    accessibility: KeychainAccessibility.first_unlock_this_device,
  ),
);

Two choices worth understanding.

first_unlock_this_device means the value is readable after the first unlock following a reboot, and — the important half — is not included in iCloud Keychain or device backups. A token that syncs to a second device is a token that outlives the device the user thinks they revoked.

encryptedSharedPreferences: true on Android uses the Keystore-backed implementation rather than plain shared preferences. Without it, “secure” storage on some configurations is considerably less secure than the name implies.

The uninstall asymmetry

A behaviour that produces confusing bug reports: on Android, uninstalling clears app storage. On iOS, Keychain items can survive an app uninstall and reinstall. A user who deletes the app to “sign out and start fresh” may reinstall and find themselves still logged in, or worse, holding a token for an account they no longer own on a resold device.

The fix is a first-run marker in ordinary preferences, which do get cleared:

Future<void> clearSecureStorageOnFirstRun() async {
  final prefs = await SharedPreferences.getInstance();
  if (prefs.getBool('has_run_before') ?? false) return;

  await storage.deleteAll();
  await prefs.setBool('has_run_before', true);
}

Call it before reading any token. It costs one read on every launch and removes an entire class of “impossible” session bugs.

Tokens: short-lived, refreshed, revocable

The storage is the easy half. The design that makes storage survivable is short access-token lifetimes plus server-side revocation:

final class AuthInterceptor extends Interceptor {
  AuthInterceptor(this._store, this._api);

  final TokenStore _store;
  final AuthApi _api;

  Future<void>? _refreshInFlight;

  @override
  Future<void> onError(DioException err, ErrorInterceptorHandler handler) async {
    if (err.response?.statusCode != 401) return handler.next(err);

    // Collapse concurrent refreshes into one.
    _refreshInFlight ??= _doRefresh();
    try {
      await _refreshInFlight;
    } finally {
      _refreshInFlight = null;
    }

    final token = await _store.readAccess();
    if (token == null) return handler.next(err);

    final retried = await _api.retry(err.requestOptions, token);
    handler.resolve(retried);
  }

  Future<void> _doRefresh() async {
    final refresh = await _store.readRefresh();
    if (refresh == null) return;
    final tokens = await _api.refresh(refresh);
    await _store.save(access: tokens.access, refresh: tokens.refresh);
  }
}

The _refreshInFlight collapsing is not an optimisation. Without it, five parallel requests hitting a 401 trigger five refreshes; if the server rotates refresh tokens on use, four of them fail and the user is logged out. This is one of the most common real-world auth bugs in mobile apps.

For login itself, RFC 8252 is unambiguous: use the system browser with PKCE, not an embedded webview. An embedded webview can read the user’s credentials, which is exactly why identity providers increasingly refuse to render in one.

What pinning does and does not buy

Certificate pinning stops an attacker with a trusted-CA-installed device from reading your traffic. It does not stop an attacker who controls the device — they can patch the check out.

final client = HttpClient(
  context: SecurityContext(withTrustedRoots: false)
    ..setTrustedCertificatesBytes(pemBytes),
);

If you pin, pin to an intermediate or a public key rather than a leaf certificate, ship at least one backup pin, and have a remote kill switch. A pinned app whose certificate rotates unexpectedly is an app that stops working for every user simultaneously, and the fix requires a store review. I have seen that outage; it is worse than the threat pinning addressed.

FAQ

Does obfuscation (--obfuscate --split-debug-info) protect my keys?

It renames identifiers. String literals remain string literals. It raises reverse-engineering cost slightly and protects nothing on its own.

Is flutter_secure_storage enough for offline-first apps with sensitive local data?

Store the encryption key there, and encrypt the database with it — for example SQLCipher via sqflite_sqlcipher. Putting megabytes of records into Keychain is not what it is for.

How do I keep a third-party key out of the app but still call the service?

Proxy through your backend. The app authenticates to you; you authenticate to them. This also gives you rate limiting and the ability to rotate the key without a release.

Can I detect a rooted or jailbroken device?

Partially, and it is an arms race. Treat it as a signal for risk scoring on your server, never as a client-side gate you rely on.

What about biometric-gated storage?

IOSOptions and AndroidOptions support requiring user presence. It is a real improvement for high-value actions, and a real friction cost — apply it to the payment confirmation, not to every app launch.

---

The Keychain and Keystore behaviours, flutter_secure_storage options, RFC 8252’s guidance on native-app OAuth, and SecurityContext API described here are documented in the references linked above. The where-secrets-belong table, the first-run cleanup pattern, the refresh-collapsing interceptor and the caution about pinning outages are my own judgement from shipping and debugging these systems. Security defaults change between package and OS versions — verify the options against the versions you actually depend on.

---

Originally published on FlutterCook. Read the latest version there — that copy is the one kept up to date.