Apple dials down “agentic” AI while upgrading Siri

Apple’s WWDC – At WWDC 2026, Apple barely touched the industry’s newest buzzword—agentic AI—spending most of the keynote instead on a more practical, command-and-response Siri AI. The shift comes with guardrails Apple says are rooted in privacy, but some of the new features

By the time the tech world hit Google I/O. Microsoft Build and NVIDIA’s Computex with “agentic” AI. the term had started to feel like a drinking game—everyone saying it. nobody letting anyone else win. Agentic AI. after all. is supposed to work on your behalf without direct input: adding meetings from emails. handling tasks autonomously. letting the AI “take the wheel.”.

Apple’s WWDC 2026 keynote this week didn’t play that game. Agentic AI was only briefly mentioned. and the company devoted most of its time to what its upgraded Siri AI can do when it’s still basically responding to you. The examples Apple highlighted weren’t sci-fi errands; they were messier. real-life annoyances—finding a friend’s new address buried in a long text message thread. or figuring out how to get tickets to an exclusive concert. In most cases, Siri AI responds to commands, with the benefit of newer AI models to better synthesize information.

Apple’s privacy story ran alongside the product pitch. With Private Cloud Compute. the company says it only uploads relevant data. makes it anonymous. and doesn’t track server logs. For the AFM3 Cloud Pro model—running on Google’s servers with NVIDIA-powered hardware—Apple says it has upgraded Private Cloud Compute to offer a similar level of security.

Craig Federighi. Apple’s SVP of software engineering. addressed the privacy architecture during a WWDC panel. explaining that PCC is an efficiency measure: “While we absolutely minimize what is sent up to PCC. the critical thing about PCC is. architecturally. that’s at that point an efficiency measure. ” he said. “Because PCC itself. by design from the ground up. is going to vaporize any record of that data the moment after it answers your question… This is not stored. It’s all in a form where it’s completely transient.”.

Based on early hands-on looks at Siri AI at WWDC 2026. it appears to work as advertised—but the article notes that long-term testing is still needed to see whether it “really holds up.” It also comes with caveats: this is the first developer beta. meaning changes could land before Apple’s latest OS updates roll out this fall.

Still, Apple couldn’t resist the momentum around agentic behavior entirely. One of the most sensitive new Apple Intelligence features is tied to passwords: the ability to automatically change compromised passwords. Apple’s Passwords app is described as “securely navigates through websites to sign in and upgrade their accounts to strong passwords.”.

That promise comes with the kind of uncertainty that keeps security teams awake. The problem isn’t that better passwords are bad—it’s what the system might do while it’s already inside accounts. The concerns raised are direct: what else could happen when Apple Intelligence logs into a site. and whether you can trust the password it comes up with. At the same time. the counterpoint is equally human: many users already have compromised passwords on less important websites. so automatically fixing them may outweigh a small loss of agency.

Apple’s new agentic touch also shows up in Safari. The browser’s updated “Notify Me” feature lets users set an alert for specific changes on a website. The use case is straightforward—tracking things like price changes or news updates for an organization—without leaving a tab open and refreshing endlessly. The article also points to where this could go next: expanding the behavior until AI could automatically purchase a hot item before it sells out. or email a group of friends if you’re watching pricing for a vacation property.

The tension here lands right in the gap between what Apple promised and what users fear. The keynote’s emphasis was on usefulness—helping Siri interpret and act on requests—while Apple framed its compute approach as private by design. Yet the same umbrella of “intelligence” still includes actions that reach outward, including sign-ins and automated updates.

It’s also hard to ignore where Apple sits in the competitive race. The article argues that Apple is behind when it comes to AI overall. pointing to how Microsoft and Google pushed AI features into consumer products quickly—sometimes in ways that arguably made those products worse for users. even if it pleased investors. Apple, it says, struggled with delays for its AI-powered Siri upgrade, first announced two years ago. Early Apple Intelligence features—notification summaries and Genmojis—were described as underwhelming, with notification summaries even leading to embarrassing mistakes.

That history matters because Apple typically approaches new technology with a specific strategy: it doesn’t always arrive first. but it tends to focus on fixing glaring issues from early entrants. making products more usable for consumers. then charging a premium for the refinement. The article references how the iPod succeeded by pairing huge music storage with easy use and iTunes as a legal digital music pathway; how the iPhone won by leapfrogging keyboard-bound smartphones with a larger screen and more usable apps.

So the hope around Siri AI isn’t that Apple will never move toward agentic systems. It’s that it will choose the moments carefully—letting models improve and users grow accustomed. while applying agentic features with judgment. The worry, stated plainly, is that the leap from “helpful assistant” to “hands-on operator” is where things can go wrong. If Siri mishears voice commands, the stakes already feel personal. The article’s closing fear is sharper: it doesn’t want to worry about Siri AI clearing out a bank account.

Apple WWDC 2026 Siri AI agentic AI Private Cloud Compute Craig Federighi AFM3 Cloud Pro Apple Intelligence Passwords app Safari Notify Me