Apple shipped a rebuilt Siri powered by Google's Gemini models after multiple years of delays. The new assistant handles multi-step requests and understands on-screen context better than its predecessor, but early testers found it still produces hallucinations and struggles with personal information retrieval.
The revamped Siri runs on a hybrid architecture. Some processing happens directly on the device. Other tasks route through Apple's Private Cloud Compute infrastructure, which Apple positions as privacy-preserving. This dual approach lets Apple balance responsiveness with the computational demands of advanced AI reasoning.
Early feedback shows genuine improvements over Siri's outdated voice assistant. Users report the new system can process complex, multi-part requests that previously would have failed. Screen understanding means Siri can see what's on your display and act on it more intelligently. An example workflow: asking Siri to find a photo from a text message and set it as your wallpaper now works reliably.
However, limitations persist. Testers documented instances where Siri hallucinates responses, fabricating information when uncertain. Access to personal context remains spotty. The system struggles to integrate knowledge from your email, calendar, and personal files consistently, despite Apple's emphasis on on-device processing enabling privacy-respecting personalization.
The Gemini partnership marks a significant shift. Apple typically builds its own AI stack, but Siri's years-long stagnation forced a reckoning. Rather than continue with outdated language models, Apple licensed Gemini from Google. This decision sidesteps Apple's ability to train competitive large language models in-house, at least for now. The deal positions Google's models as the backbone of Apple's assistant strategy across devices.
Geographic restrictions complicate the rollout. European Union regulators have blocked Siri's availability in the region. This reflects broader friction between Apple's AI plans and EU Digital Markets Act requirements. The bloc views Apple's ability to limit third-party access to certain features as anticompetitive. Siri's unavailability in the EU stems partly from these regulatory pressures and partly from technical complications around data localization rules.
For users outside the EU, the rebuilt Siri represents the most substantial improvement to Apple's assistant in years. The system addresses the core failure of previous versions: inability to handle real-world queries that require reasoning across multiple steps. Screen context awareness tackles another chronic weakness. Voice control integration works more naturally when Siri actually understands what you're looking at.
The hallucination problem requires attention. Any AI system generating false information with confidence undermines user trust. Apple will need to improve Gemini's reliability on device-specific tasks. Personal context gaps suggest the hybrid on-device and cloud approach may not yet achieve Apple's privacy promises. If personal data needs cloud processing to power personalization, the privacy advantage over pure cloud models narrows.
Apple's reliance on Gemini creates dependency on Google's continued investment and reliability. Any performance degradation or pricing changes to Gemini access would ripple through Apple's entire assistant strategy. This partnership also raises questions about how long Apple remains comfortable outsourcing core AI capabilities rather than building internally.
The EU exclusion signals that regulatory complexity will shape AI assistant adoption. Where regulations block deployment, users lose access to improvements. This dynamic will likely continue as different jurisdictions impose conflicting requirements on AI systems.