iOS 27 brings Siri AI in CarPlay with meaningful context memory and natural conversation, alongside new audio scrubbing in Now Playing, a persistent MiniPlayer, and improved wireless reliability, while Maps gains natural language route search that can leverage Siri AI context in the future.

The CarPlay upgrade features an on-device, three-tier inference architecture that combines an on-device 3 billion-parameter model for simple tasks, Private Cloud Compute for intermediate work, and a licensed Google Gemini model for advanced reasoning, with a system orchestrator routing tasks accordingly.

The architecture relies on an orchestrated mix of on-device, private cloud, and licensed AI, enabling a balance between responsiveness and capability.

Siri AI isn’t available in the EU at launch due to Digital Markets Act negotiations and isn’t available in China pending regulatory approval, with EU Mac and Vision Pro users able to access it when language-appropriate.

Third-party app integration remains limited for now; a new conversation mode API could enable it later, but it’s not delivered in practice yet.

A major current limitation is lack of live traffic data access; Siri AI cannot pull real-time traffic information directly and relies on indirect sources like news or social posts about road incidents.

Video streaming via AirPlay is introduced for CarPlay, but it’s restricted to parked use with safety restrictions, and device certification through Apple’s MFi Program is required.

On-device Siri AI is available only on iPhone 15 Pro and newer for the on-device tier, while iOS 27 features work on older iPhones for non-Siri AI CarPlay enhancements, with public beta planned mid-July and full release expected in September 2026.

For frequent CarPlay users with iPhone 15 Pro or newer, iOS 27 marks the most significant CarPlay upgrade in 12 years, though expectations should be tempered around live traffic access, third-party app integration, and mid-drive interruption handling.

In real-world testing, Siri AI demonstrates memory and organization of notes across minutes, applies context to multi-step requests, reduces taps and interruptions, and delivers a more human-like tone.