Siri
Last reviewed
Apr 28, 2026
Sources
37 citations
Review status
Source-backed
Revision
v1 ยท 4,239 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Apr 28, 2026
Sources
37 citations
Review status
Source-backed
Revision
v1 ยท 4,239 words
Add missing citations, update stale details, or suggest a clearer explanation.
| Siri | |
|---|---|
| Type | Virtual assistant, conversational AI |
| Developer | Apple Inc. |
| Initial release | October 14, 2011 (with iPhone 4S) |
| Predecessor | Siri Inc. (iOS app, February 2010) |
| Origin | SRI International spin-out (2007) of the CALO project |
| Platforms | iOS, iPadOS, macOS, watchOS, tvOS, visionOS, audioOS (HomePod) |
| Languages | 21 languages across roughly 36 countries (as of 2024) [1] |
| Key people | Adam Cheyer, Tom Gruber, Dag Kittlaus (founders); Steve Jobs (Apple acquirer); John Giannandrea (current AI/ML lead) |
| Website | apple.com/siri |
Siri is a virtual assistant developed by Apple Inc. and integrated into iPhone, iPad, Mac, Apple Watch, Apple TV, Vision Pro, and HomePod. It accepts voice and text input, interprets natural language requests, executes tasks across system apps and supported third-party apps, and replies through synthesized speech or on-screen results. Siri launched on the iPhone 4S on October 14, 2011, making it the first mainstream voice assistant shipped on a smartphone [2][3].
The product traces back to the CALO project, a DARPA-funded artificial intelligence research effort led by SRI International between 2003 and 2008 [4][5]. SRI spun the work into a startup called Siri Inc. in 2007, founded by Adam Cheyer, Tom Gruber, and Dag Kittlaus. The company shipped a standalone iOS app in February 2010 and was acquired by Apple two months later for a sum widely reported as more than 200 million US dollars [6][7]. Apple killed the standalone app, halted plans for Android and BlackBerry versions, and rebuilt Siri as an operating system feature.
For most of its first decade Siri positioned itself as a hands-free assistant for short tasks: setting alarms, sending messages, looking up weather, opening apps, and controlling music playback. Apple expanded its capabilities over many releases, adding the always-listening "Hey Siri" trigger in iOS 8 (2014), third-party developer integration through SiriKit in iOS 10 (2016), the user-programmable Shortcuts app in iOS 12 (2018), and on-device speech recognition in iOS 15 (2021) [8][9][10][11]. Siri received persistent criticism throughout these years for limited contextual understanding, weaker accuracy than rivals, and slow adoption of large language model techniques.
In June 2024 Apple announced Apple Intelligence, a generative AI system that included a substantial Siri overhaul. The first wave shipped with iOS 18.1 in October 2024; the much-promoted personalized Siri features were delayed and, by mid-2025, pushed to a target of spring 2026 [12][13][14].
Siri's intellectual lineage is the CALO project, short for Cognitive Assistant that Learns and Organizes. CALO was funded by the DARPA Personalized Assistant that Learns (PAL) program and ran from May 2003 through 2008 under SRI International as the lead contractor [4][5]. SRI coordinated more than 300 researchers across roughly 22 universities and research labs, including Carnegie Mellon, Stanford, the University of Massachusetts at Amherst, and Yale. The published budget figure was approximately 150 million US dollars over five years [4]. Some popular accounts have cited a 200 million dollar figure that includes related sub-programs and follow-on work; the SRI primary source describes the headline figure as 150 million [4].
CALO's research goal was to build a system that could reason, learn from experience, take action under instruction, explain its decisions, reflect on outcomes, and adapt to surprise. The work pulled together planning, machine learning, dialogue management, speech recognition, and natural language understanding into a single integrated assistant. Among CALO's spin-outs were the open-source SPARK task scheduler, the Active web framework, and Siri itself. SRI's history pages describe Siri as "perhaps the most well-known" of the program's commercial outgrowths [5].
The name Siri is sometimes attributed to a Norwegian girl's name meaning "beautiful woman who leads you to victory." Co-founder Dag Kittlaus, who is Norwegian, has confirmed that interpretation in interviews [15]. The acronym treatment, sometimes given as "Speech Interpretation and Recognition Interface," appeared in coverage but was not the original derivation.
In 2007 SRI authorized a commercial spin-out called Siri Inc. The three founders were:
Siri Inc. raised about 24 million US dollars across two rounds from Menlo Ventures, Morgenthaler Ventures, and SRI itself, with later participation from Horizons Ventures [15][16]. The team built a consumer-facing product around the same do-engine architecture used in CALO: parse user intent, route the request to a domain-specific handler, call partner web services for data, and stitch the result back into a conversational reply.
The Siri app shipped on the App Store on February 4, 2010 [17]. The first release supported a narrow set of practical tasks. Users could ask Siri to find a restaurant and book a table through OpenTable, look up movie times and buy tickets through MovieTickets.com, hail a cab through Taxi Magic, search local businesses through Yelp, and retrieve weather and traffic. The app was free and ran only on iPhone and iPod touch. Versions for Android and BlackBerry were in active development but never shipped publicly.
Speech recognition for the original app was licensed from Nuance Communications, the dominant supplier of voice technology at the time [18]. Apple did not publicly acknowledge the Nuance dependency for years; Nuance chief executive Paul Ricci confirmed the relationship at a 2013 industry conference [18][19]. Apple progressively reduced its reliance on Nuance, eventually replacing the speech stack with in-house models. Microsoft acquired Nuance in 2021 for about 19.7 billion US dollars, by which point the link to Siri's day-to-day operation had largely been severed [20].
Steve Jobs personally pursued Siri Inc. within weeks of the app's launch. According to Cheyer and Kittlaus, Jobs called the company headquarters directly and arranged a meeting at his Palo Alto home. After roughly three weeks of negotiations, Apple announced its acquisition of Siri Inc. on April 28, 2010 [6][7]. Apple has never disclosed the price; press coverage at the time, citing sources close to the deal, reported a value above 200 million US dollars [6][16].
After the deal closed, Apple pulled the Siri iOS app from the App Store on October 15, 2011, the day after the iPhone 4S launch. The pre-existing Android and BlackBerry projects were canceled. Cheyer became director of engineering inside Apple's iPhone group, Kittlaus became director of iPhone apps, and Gruber stayed on as design lead for the assistant's voice and persona.
Apple announced the iPhone 4S on October 4, 2011 at its Cupertino campus, with Siri as the headline feature. Steve Jobs, who had stepped down as chief executive on August 24, 2011, watched a private demo of the launch product the day before but did not attend the event itself. Jobs died on October 5, 2011, the day after the announcement [21]. The iPhone 4S, and Siri with it, went on sale in the United States, the United Kingdom, Canada, Australia, France, Germany, and Japan on October 14, 2011 [3]. Apple shipped Siri labeled as a "beta" feature, exclusive to the 4S, and reviewers praised its conversational tone while flagging slow response times and frequent misunderstandings.
Siri's high-level pipeline has remained recognizable across its evolution. A typical voice request flows through five stages.
From 2024 onward, Apple Intelligence added a generative layer running alongside this pipeline. Requests that benefit from open-ended language generation (rewriting a paragraph, summarizing a long email, answering general knowledge questions) flow through Apple's foundation models or, with user consent, are handed off to ChatGPT. Routine commands continue to use the legacy intent system, which is faster and more predictable for high-volume actions like setting timers and sending messages.
| Year | iOS / OS release | Key Siri changes |
|---|---|---|
| 2011 | iOS 5 (iPhone 4S) | Initial launch as a beta feature, English (US, UK, Australia) [3] |
| 2012 | iOS 6 | Sports scores, restaurant info, movie times, app launching, French, German, Japanese, and additional locales |
| 2012 | iOS 6 | iPad mini and iPad 3 support |
| 2013 | iOS 7 | New male and female voices in select languages, expanded data sources, Twitter and Wikipedia integration |
| 2014 | iOS 8 | "Hey Siri" wake phrase added, requiring the device to be plugged in [8] |
| 2015 | iOS 9 | Proactive suggestions in Spotlight, contextual reminders, deeper search [24] |
| 2016 | iOS 10, macOS Sierra | SiriKit opens to third-party apps in defined domains; Siri arrives on the Mac [9][25] |
| 2017 | iOS 11 | Hybrid deep-learning text-to-speech voices with more natural prosody [23] |
| 2017 | HomePod (December 2017 announce, February 2018 ship) | Siri inside Apple's first smart speaker |
| 2018 | iOS 12 | Shortcuts app, custom voice triggers for app actions [10] |
| 2019 | iOS 13 | New Neural TTS voices, audio handoff to AirPods, more on-device handling for some queries; Voice Control accessibility feature added |
| 2020 | iOS 14 | Compact UI that no longer takes over the full screen, on-device dictation expanded |
| 2021 | iOS 15 | On-device speech recognition for many commands without internet on A12 Bionic and newer; expanded sharing intents [11] |
| 2022 | iOS 16 | Faster activation, automatic message send option, expanded language pairings |
| 2023 | iOS 17 | Trigger word shortened to just "Siri" on supported devices, back-to-back requests without re-triggering |
| 2024 | iOS 18.1 (October 2024) | First wave of Apple Intelligence: redesigned glow UI, type-to-Siri, richer language understanding [13] |
| 2024 | iOS 18.2 (December 2024) | ChatGPT handoff with explicit user consent, Genmoji, Image Playground [26] |
| 2025 | iOS 18 cycle | Personalized Siri (on-screen awareness, in-app actions, personal context) delayed; reset to spring 2026 [12][14] |
Apple announced Apple Intelligence at WWDC on June 10, 2024 [27]. The system is the most ambitious refresh of Siri since the iPhone 4S launch and adds a layer of generative capability across system apps. Apple frames the work in terms of three model tiers:
Apple Intelligence works only on the iPhone 15 Pro, iPhone 15 Pro Max, and the entire iPhone 16 line, plus Macs and iPads with M1 or later chips. The hardware requirement is driven by memory and neural-engine demands. The system rolled out in waves: iOS 18.1 in October 2024 introduced writing tools, notification summaries, and a redesigned Siri glow at the screen edge; iOS 18.2 in December 2024 added Genmoji, Image Playground, and the ChatGPT integration [26][13].
The most-anticipated part of the announcement was a personalized Siri capable of acting across apps, drawing on personal context like messages and calendar entries, and seeing what is on screen. Bloomberg's Mark Gurman reported in February 2025 that the feature had hit serious engineering issues. Apple confirmed an indefinite delay in March 2025, with Siri chief Robby Walker reportedly calling the situation "ugly" and "embarrassing" in a staff meeting. By June 2025 Apple had reset the internal target to spring 2026 [12][13][14]. Reporting also indicated Apple had been working on two parallel Siri backends, an older intent system and a newer LLM-based one, which complicated integration [13].
The internal foundation model project at Apple was reported by Bloomberg in 2023 under the codename "Ajax," with the chatbot front end nicknamed "Apple GPT." Apple confirmed parts of the work formally at WWDC 2024 without publicly using those internal names.
The public partnership with OpenAI was announced jointly with the Apple Intelligence reveal. Apple stated that no payment changes hands between the two companies for the integration; the value to OpenAI is exposure and the chance to surface ChatGPT subscriptions to Apple users, while the value to Apple is delivering frontier model quality without owning a frontier model [30].
Apple has framed Siri's privacy posture as differentiated from that of cloud-first rivals. Three concrete practices reflect this position.
On-device processing. Wake-word detection, the bulk of dictation, and many simple requests run on the device rather than in the cloud. iOS 15 made this a major selling point in 2021, advertising offline support for many actions on iPhones with A12 Bionic or later chips [11]. Apple Intelligence extends the pattern: the on-device foundation model handles tasks within its capability, and only requests that need server compute are routed to Private Cloud Compute [29].
Differential privacy. Where Apple does collect aggregate data to improve Siri (for example, the popularity of a misheard phrase across many users), it applies differential privacy noise to the data on device before transmission, limiting what can be inferred about a single individual.
Private Cloud Compute. When a request must leave the device, Apple Intelligence routes it to a server stack the company designed specifically for AI workloads. Apple states that data is used only to fulfill the request, that the servers do not retain data after use, and that independent researchers can audit the production builds. The architecture pairs verifiable cryptographic attestation with a hardened software stack to enforce these properties [29].
The most public stress test of Apple's privacy framing was the 2019 Siri grading controversy. On July 26, 2019, The Guardian reported that Apple contractors hired to review a small fraction of Siri recordings for accuracy regularly heard private conversations, including medical details, drug deals, and sexual encounters. The report, based on a whistleblower account, prompted immediate criticism. Apple suspended the global grading program on August 2, 2019, while it reviewed the practice [31][32]. When Apple reinstated quality review later that month, it made participation opt-in for users, deleted historical recordings linked to the program, and shifted review work to Apple employees rather than third-party contractors [33].
In 2025 Apple agreed to a 95 million dollar settlement of a class action lawsuit alleging that Siri had captured audio without user consent. Apple denied wrongdoing and said the settlement avoided continued litigation costs.
The iPhone 4S Siri launch was met with broad public excitement and a strong marketing push from Apple, including a series of celebrity television advertisements with John Malkovich, Samuel L. Jackson, and Zooey Deschanel in 2012. Reviewers were less unanimous: many noted impressive natural language understanding for the era while flagging slow servers, missed activations, and a narrow set of supported tasks. Apple's "beta" label set expectations accordingly.
Over the following decade Siri became a recurring punchline in coverage of Apple's AI work. Common criticisms included:
Satisfaction surveys, including Loup Ventures' annual smart speaker IQ test from 2017 to 2019, consistently placed Siri behind Google Assistant on raw question-answering accuracy.
Leadership inside Apple's AI work has churned visibly since the original Siri team integrated in 2010.
Bloomberg's reporting through 2023, 2024, and 2025 has described internal disagreements over how aggressively Apple should pivot to LLM-based architectures, friction between the Siri team and the central foundation models team led by Giannandrea, and frustration among engineers over shipping deadlines that slipped after WWDC 2024 [12][13][14]. In response, Apple reorganized parts of the AI group in 2025 and reassigned Vision Pro lead Mike Rockwell to oversee the Siri team while Giannandrea retained the broader AI mandate.
| Assistant | Developer | Initial release | Primary surfaces | Foundation models |
|---|---|---|---|---|
| Siri | Apple | October 2011 (iPhone 4S) | iPhone, iPad, Mac, Apple Watch, Apple TV, HomePod, Vision Pro | Apple ~3B on-device + Apple server + optional ChatGPT [28][30] |
| Google Assistant | May 2016 (Allo, Pixel) | Android, Google Home, Pixel, Wear OS, Android Auto | Google PaLM 2, Gemini (replacing Assistant on phones from 2024) | |
| Alexa | Amazon | November 2014 (Echo) | Echo speakers, Fire TV, Fire tablets, third-party devices | Amazon Titan and Nova, Alexa+ generative tier from 2025 |
| Cortana | Microsoft | April 2014 (Windows Phone 8.1) | Windows, Microsoft 365 (consumer Cortana retired 2023) | Replaced for consumers by Copilot |
| Bixby | Samsung | March 2017 (Galaxy S8) | Samsung phones, tablets, TVs, smart home | Built atop Viv technology Samsung acquired in 2016 |
| ChatGPT voice | OpenAI | September 2023 (voice mode), May 2024 (Advanced Voice with GPT-4o) | iOS, Android, web | GPT-4o and successors |
Siri's footprint is broader than any rival in terms of platforms (iPhone alone shipped over 230 million units in 2024) and its integration depth across Apple silicon is unique. Rivals have generally led on raw model quality, breadth of skills or actions, and openness to third-party developers.
Siri at launch supported only English (US, UK, and Australian variants). Apple has expanded over the years through annual additions. By 2024, Apple's official documentation listed support for 21 base languages spanning roughly 36 country locales, including English (eight regional variants), Spanish (US, Mexico, Spain, Chile), French (France, Canada, Switzerland, Belgium), German, Italian, Mandarin Chinese (mainland and Taiwan), Cantonese, Japanese, Korean, Russian, Arabic, Hebrew, Thai, Turkish, Danish, Swedish, Norwegian Bokmal, Finnish, Dutch (Netherlands and Belgium), Portuguese (Brazil and Portugal), and Malay [1].
Apple Intelligence launched with US English only in October 2024, then expanded to localized English (UK, Australia, Canada, Ireland, New Zealand, South Africa) in December 2024 (iOS 18.2). Additional languages, including French, German, Italian, Portuguese, Spanish, Japanese, Korean, and Chinese (Simplified), were added in 2025 [13][26].