iOS 18.5 privacy tools expand with on-device model auditing

AI · 3 min read

With iOS 18.5, Apple has added a new Privacy Audit feature that generates human-readable explanations of on-device model behavior for apps that use Core ML. The summaries highlight which input factors influenced a prediction and provide a confidence score, all computed locally to avoid network exposure.

Developers are required to include an 'explainability manifest' when submitting apps that use certain classes of models. Apple’s App Review will sample manifests to ensure claims about transparency are accurate and that explanations do not leak private inputs. The company says this aims to reduce opaque personalization in sensitive categories such as health and finance.

Designers will need to craft explanation UIs that are both concise and informative; Apple includes new Human Interface Guidelines for presenting provenance, confidence, and recourse actions. The update aligns with Apple’s privacy-first branding while nudging platforms toward explainable ML in consumer devices.