Apple is fundamentally rebuilding its Siri voice assistant using an architecture based entirely on large language models (LLMs). This major strategic pivot follows significant internal challenges. Apple’s initial “Apple Intelligence” features have reportedly faltered, according to a Bloomberg report. Attempts to merge generative AI with Siri’s old framework were described by one source as a “wreck”. The overhaul signals a critical moment for Apple as it aims to deliver a more intelligent, conversational Siri and catch up with competitors like Google’s Gemini and OpenAI’s ChatGPT.
The decision to reconstruct Siri stems from several deep-rooted issues, including a late entry into the generative AI field. Some Apple leaders, like software chief Craig Federighi, were reportedly hesitant about large AI investments without clear product outcomes. Merging new AI with Siri’s existing, aging infrastructure also proved difficult; an Apple employee informed Bloomberg that fixing one issue often caused three more to appear. Furthermore, Apple’s AI chief, John Giannandrea, reportedly underestimated the demand for AI chatbots and did not aggressively pursue necessary funding or view competitors as an immediate threat. Marketing promises also outpaced development, leading to delays for anticipated features.
Consequently, Apple’s Zurich-based AI team is now developing a new Siri architecture. Bloombergs’s Mark Gurman reported this new system will be built on an LLM-based engine, aiming to make the assistant more believably conversational and better at synthesizing information. This strategic redirection coincides with John Giannandrea reportedly being removed from Siri and other product development roles in spring 2025. Some Apple executives discussed putting him on a path to retirement, though Gurman noted Giannandrea intends to stay, saying “relieved Siri is now someone else’s problem.”
A Pattern of Delays and AI Missteps
Apple’s journey toward an enhanced Siri has been marked by persistent delays and public setbacks, illustrating the company’s struggle to keep pace with AI’s rapid evolution. Reports emerged as early as November 2024 that a fully revamped “LLM Siri” might not launch until 2026. This was despite Apple’s unveiling of “Apple Intelligence” and initial Siri upgrades at WWDC 2024, which some observers deemed largely superficial.
Initial plans to integrate OpenAI’s ChatGPT into Siri, first expected with iOS 18 in autumn 2024, also faced postponements. Last summer it became evident that these key Siri improvements would likely be delayed until spring 2025, potentially arriving with iOS 18.4. Then, by March 2025, Apple officially acknowledged these challenges. An Apple spokesperson confessed, “It’s going to take us longer than we thought to deliver on these features, and we anticipate rolling them out in the coming year.” Internal iOS 18.4 tests in early 2025 found key features non-functional, with demos allegedly using pre-recorded videos, leading to an indefinite postponement.
Beyond timelines, Apple’s AI initiatives have encountered significant criticism. Leaked internal evaluations from October 2024 indicated Siri’s accuracy was considerably lower than OpenAI’s ChatGPT. Apple Intelligence itself faced scrutiny for disseminating false news headlines within iOS.
Adding to the problems, Apple also had to agree to a $95 million settlement concerning allegations that Siri had improperly recorded user conversations. Functional glitches also plagued the assistants interplay with Apple Intelligence.
One notable example occurred in December 2024 when Siri’s proactive AI mistakenly created phantom restaurant reservations for users who had only browsed OpenTable.
Forging a New Path Amidst Competitive Pressure
Apple’s current strategy involves more than just integrating LLMs; it’s a fundamental reconstruction of Siri. The company is exploring using on-device processing and differential privacy to enhance AI training data. This method involves comparing synthetic data with anonymized language patterns from user emails, all while maintaining Apple’s privacy commitments.
Mark Gurman also reports that Apple is considering allowing the new LLM-Siri to access and synthesize data from various web sources, similar to Perplexity, which in April launched its own AI voice assistant. Apple has reportedly engaged in partnership discussions with Perplexity for AI search functionalities within the Safari web browser.
Bloomberg’s Gurman says that Apple intends to market Siri separately from Apple Intelligence, possibly to shield its broader AI branding from Siri’s tarnished reputation. For European Union residents, Apple is also reportedly developing a feature to allow complete replacement of Siri with third-party voice assistants.
This extensive overhaul is taking place as Apple’s competitors solidify their positions in the AI assistant space. Google’s Gemini Live already launched in August 2024, and since then is being deeply integrated in Android and Google’s whole ecosystem, replacing Google Assistant. In May, Google launched Gemini Live for Workspace accounts, introducing interactive voice conversations, screen sharing, and camera integration to its AI assistant.
Navigating Privacy and Market Realities
Amazon, suffering from similar delays with its “Project Banyan” AI overhaul for its Alexa assistant, has made substantial progress thanks to its partnership with Anthropic for using Claude AI. The company finally introduced a more sophisticated Alexa+ in early 2025 which offers conversational memory and real-time automation.
Apple’s long-held strategy of prioritizing on-device processing for user privacy, a key differentiator, appears to be a constraint for the ongoing Siri overhaul. This approach may limit Siri’s capacity to learn from the extensive datasets that fuel cloud-based AI systems, presenting a significant hurdle as Apple endeavors to transform its iconic assistant into a truly intelligent and competitive tool. As Siri news are currently mainly bad ones, Gurman suggests that Siri upgrades are unlikely to be a central focus at WWDC 2025, with Apple expected to concentrate on iOS 19 instead.
