Cupertino promised magic this spring; the people building it are bracing for something else. If the iPhone must choose between privacy and prowess, which side will users notice first?
Inside Apple, the grand spring makeover for Siri is colliding with the messier reality of bugs, slow responses, and missing tricks. Employees say the assistant sometimes punts tasks to ChatGPT, while Cupertino considers slapping a beta label on features that were meant to shine in iOS 26.4. The company credits its privacy-first push with Linwood’s on-device processing for the holdup, even as it readies a more ambitious chatbot, Campo, for later. With eyes now drifting to iOS 27 in the fall, the gap between promise and product is where the real story sits.
High hopes, unmet expectations
When Apple teased sweeping AI updates for Siri to land with iOS 26.4 this spring, the mood among users and developers was buoyant. Internal voices now suggest those ambitions met tougher realities. Instead of the sleek leap promised onstage, the latest Siri builds confront bugs, performance dips, and features that feel unfinished. Was the hype ever realistic?
The challenges with Siri on iOS 26.4
The assistant’s day-to-day performance still frustrates. Users report lagging replies, failed lookups, and commands that stall midstream. This is the case even with simple requests that should be handled in milliseconds.
Delayed responses and timeouts
Missed or partial data retrieval
Interrupted voice capture for fast speakers
Occasional handoff to external AI, like ChatGPT
According to people familiar with testing, several marquee upgrades originally tied to iOS 26.4 are sliding to iOS 27 (expected this fall). There has even been discussion of flagging certain features as Beta to temper expectations. In addition to that, limited scenarios reportedly trigger a fallback to third-party models, breaking the seamlessness users expect from Siri.
The balance between innovation and privacy
One reason for the slowdown sits at the heart of Apple’s approach: a privacy-first architecture. Rather than stream every prompt to public clouds, Apple leans on its “Linwood” system to process requests locally or on tightly controlled private compute (on-device when possible). That stance reduces data exposure but raises formidable engineering hurdles.
Craig Federighi has repeatedly underlined that user queries should remain private and not feed broad training pipelines. The trade-off is clear: keeping data close means building smaller, smarter models, advanced scheduling, and resource management that rivals typically offload to massive server farms. Indeed, that complexity helps explain why some Siri features feel slower to reach parity.
What’s next: the promise of iOS 27
Eyes now turn to iOS 27. Insiders point to a more capable chatbot effort, codenamed Campo, designed to tighten Siri’s grasp of context and tasks. For example, smoother cross-app actions, richer follow-up questions, and better memory of recent activity are said to be in scope. The goal: reduce handoffs, lift accuracy, and make Siri feel reliably helpful.
None of this guarantees instant redemption, but the direction is coherent. Local protections stay intact, while the assistant gains breadth and speed. If Apple can nail performance and polish—without loosening its privacy guardrails—Siri could finally step beyond timers and playlists, and start earning its spot at the center of the iPhone experience.