Apple’s first standalone Siri app, coming in iOS 27, will include an auto-delete function for chat histories that borrows from the Messages app. Users will be able to configure the app to retain conversations for 30 days, one year, or indefinitely. The feature, reported by Bloomberg’s Mark Gurman in his Power On newsletter on Sunday, is designed to position Apple’s approach to AI privacy as structurally different from competitors that offer temporary or incognito chat modes as optional settings users must manually enable.

The Siri app will function as a chatbot akin to ChatGPT or Claude, serving as a repository for past conversations that users can search, continue, or delete. It can be accessed either through the standard Siri activation, via the side button or wake word, or through a new “Search or Ask” mode triggered by swiping down from the top centre of the screen. The app will support both voice and text input, file uploads, and web-sourced answers with images and bullet points. Users will be able to choose whether the app opens to a grid of prior conversations or a new chat each time.

The privacy architecture is Apple’s competitive differentiator, and also its excuse. Competing chatbots rely heavily on conversation histories and memory systems to personalise responses and improve over time. Apple is placing tighter limits on what information can persist and how long it is retained, building the restrictions into the system itself rather than offering them as optional modes. Meta launched a temporary chat feature just last week. Apple’s position is that such protections should not require users to opt in.

The strategic context makes the privacy framing more interesting than it appears on the surface. Apple has quietly replaced much of its own AI infrastructure with Google’s Gemini, paying roughly $1 billion annually for a custom 1.2-trillion-parameter model that will power the next-generation Siri. The company’s existing partnership with OpenAI is fraying, with OpenAI’s lawyers preparing possible legal action over a ChatGPT-Siri deal that failed to deliver the subscription revenue OpenAI expected. iOS 27 will introduce a system called Extensions that allows users to install rival AI chatbots and route Siri queries through whichever model they choose, including Claude and Gemini, effectively downgrading ChatGPT from a privileged partner to one option among several.

Gurman notes that Apple has been less specific about how the new Siri infrastructure will be hosted and operated at scale. The company has said the revamped Siri will use Private Cloud Compute, its extension of the iPhone’s security model into the cloud, but has not confirmed that it will rely on the same Apple-designed chips, data centres, and security architecture as the current system. The implication is that Google’s cloud infrastructure will handle some of the workload, something Apple does not want to emphasise given the tension with its privacy messaging.

The auto-delete feature is clever positioning regardless of whether the privacy argument is fully coherent. By defaulting to structured retention limits rather than permanent storage, Apple can claim its AI assistant is designed to forget, a contrast to systems that are designed to remember everything in order to improve. Whether users will value that distinction depends on whether they notice the trade-off: a Siri that forgets your preferences after 30 days is also a Siri that cannot learn from your history the way ChatGPT or Claude can.

Perhaps the most revealing detail in Gurman’s report is that the new Siri may launch as a beta, even after a two-year delay. The revamped assistant was originally scheduled for 2024. Test versions of iOS 27 within Apple currently use a beta label for the new Siri and include a toggle allowing users to opt out. Apple is simultaneously developing AI smart glasses for 2027 that will use the same Gemini-powered Siri as their primary interface, meaning the assistant needs to work reliably across multiple form factors by next year.

Gurman frames the stakes clearly: Tim Cook does not want his final launch as CEO to be a misstep. Apple got “some slack” with the original Apple Intelligence rollout in 2024, but the competitive and regulatory landscape has shifted. Google’s Gemini has grown its web traffic share from 5.7% to 21.5%. The EU is preparing to force both Apple and Google to open their AI assistants to rivals under the Digital Markets Act. Android 17, announced at Google I/O this week, will ship with a Gemini Intelligence system and a new Googlebook laptop platform. Apple is entering this environment with a Siri that is two years late, powered by a competitor’s model, and potentially still labelled as unfinished.

The Genmoji upgrade is the lightest piece of the iOS 27 picture but speaks to the same pattern. Apple’s AI-generated emoji feature was poorly received at launch: images looked nothing like the advertisements, and the models used so much power that phones would heat up and drain their batteries. iOS 27 will add “Suggested Genmoji,” automatically generated from a user’s photos and commonly typed phrases, in an attempt to increase adoption of a feature that was supposed to demonstrate Apple Intelligence’s consumer appeal.

Apple’s privacy argument arrives at a moment when the concept has never been more salient in consumer hardware. Meta is facing lawsuits and regulatory investigations over how its Ray-Ban smart glasses handle user footage. Google’s Android XR glasses will ship with cameras and Gemini AI. Snap is launching consumer AR specs with built-in AI. In this environment, an AI assistant that auto-deletes your conversations and does not train on your data is a genuine differentiator, provided the assistant itself is good enough that people want to use it. That is the question Apple has been unable to answer for 15 years. Whether Gemini, a beta label, and a privacy toggle change the answer will be clear by September.