Apple's rebuilt Siri AI shipped in the iOS 27 developer beta on June 9, 2026, and on a first hands-on it does the basic things Apple promised two years ago and failed to deliver. The new assistant parses event lists from emails into calendar entries in a single prompt, pulls personal context from on-device data like mail and messages, and answers questions such as when to leave for the airport by cross-referencing a calendar event with a confirmation email. It runs on Gemini models under the hood, with heavier prompts routed to Apple Private Cloud Compute.
The feature set is narrow by 2026 standards. Multi-event calendar parsing from a screenshot has worked on Android via Gemini for at least a year, and plant-diagnosis and reminder chains have been live for months. Siri AI is roughly where Gemini was in 2025 — but it ships, which is the part that matters for Apple after the first iteration of Apple Intelligence's Siri overhaul never materialized as demoed at WWDC 2024.
The Verge's Allison Johnson tested the assistant by asking it to diagnose wilting roses, build a hardware-store shopping list, and set a compost reminder — all from a single conversational thread. She also asked when to return camera gear rented for WWDC; Siri pulled the due date from a calendar event and a confirming email without being told where to look.
“This is like, baby's first AI assistant stuff, but it's huge that it actually works.”— Allison Johnson, Senior reviewer
Key facts
- 01Apple's rebuilt Siri AI shipped in the iOS 27 developer beta on June 9, 2026, two years after a botched first attempt at WWDC.
- 02The new Siri runs on Gemini models, with on-device personal context routed to Apple Private Cloud Compute for heavier prompts.
- 03Siri AI can now parse calendar events from email, build shopping lists, and answer context-aware questions like airport departure timing.
- 04Gemini on Android has handled multi-event calendar parsing for at least a year and plant diagnostics for months, putting Siri behind on parity.
- 05Siri is now summoned from the Dynamic Island via long-press of the wake button, replacing the prior glowing-border treatment.
Under the hood, Apple is taking a different personal-context approach than Google. Apple indexes on-device data from mail, messages, and calendar so Siri can surface the relevant slices when needed. Prompts that exceed on-device capacity get forwarded to Apple Private Cloud Compute with only the necessary personal data attached. Gemini on Android, by contrast, asks users to opt into Gmail and calendar access and queries those sources directly at runtime.
Voice and tone are also tuned differently. Asked the same wilting-flowers question, Gemini opened with "That is incredibly frustrating…" before listing causes. Siri skipped the sympathy and went straight to diagnosis. Apple has not detailed which Gemini variant powers Siri, nor the split between on-device and cloud inference.
“New Siri is built on Gemini models, so it makes a lot of sense that the first iteration of Siri AI feels a little bit "Gemini, circa 2025."”— Allison Johnson, Senior reviewer
Guardrails held in light probing. Johnson reported that shady prompts returned a curt "I can't help you with that," though she did not run an adversarial stress test. The integration touches more of the iPhone surface than the old Siri: a glowing "search or ask" prompt sits at the top of homescreen search, and a long-press of the wake button summons Siri from the Dynamic Island rather than the prior screen-border treatment.
What Siri AI does not do is also notable. It will not place a DoorDash order, run a multi-step agentic task, or execute the cross-app actions Apple demoed at WWDC 2024 and then quietly walked back. Apple has scoped this release to features it can ship reliably — calendar parsing, list-building, context-aware Q&A — rather than re-promising the agent-style demo that damaged Siri's credibility.
That conservative scoping is the point. Apple needs to re-establish that Siri does what it says, and a beta that handles parent-grade chores like dumping a school spirit-week flyer into a calendar matters more for trust than a flashy demo. The company also continues to lean on outside partners — Gemini for the model layer, and as previously covered on this site, Nvidia Blackwell GPUs running inside Apple Private Cloud Compute on Google Cloud for the inference layer.
“Apple needs this version of Siri to earn back trust.”— Allison Johnson, Senior reviewer
The catches are real. The Gemini dependency means Apple is building its assistant strategy on a Google substrate, which has obvious strategic risk if the relationship sours or if Google decides Gemini on Android should always lead Siri on iPhone in capability. The feature gap with Android is also wide: months and years of Gemini iteration sit between today's Siri and feature parity. And the assistant is in a developer beta, not general availability — past Apple AI demos have been further along in polish than in shipping.
For Apple, the strategic read is that shipping a working baseline beats promising an agentic future. The Siri rebuild is no longer a moonshot bet on Apple's own foundation models; it is a Gemini-powered product that does a narrower set of things well, plus a privacy story about on-device indexing and Private Cloud Compute. For the broader AI market, it confirms that even Apple — the company with the most to lose by depending on a rival — has concluded that buying frontier model capability is faster than building it. Whoever owns the model layer increasingly sets the floor for what every consumer assistant can do.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
