Apple's development of the new Siri encounters further obstacles

On February 11th, local time, tech journalist Mark Gurman reported that Apple’s long-planned Siri virtual assistant upgrade has encountered issues during recent testing, potentially delaying the release of several highly anticipated features. Apple initially planned to include these new features in the upcoming iOS 26.4 update scheduled for March, but is now considering rolling them out gradually in future versions. This means that at least some features may not be available until the release of iOS 26.5 in May or iOS 27 in September.

This is the second major delay for the new Siri since its upgrade was announced in 2024. In June 2024, Apple first announced at its Worldwide Developers Conference that Siri would undergo a significant AI upgrade, leveraging its in-house large language model platform, Apple Foundations Models, to create a smarter personal voice assistant, originally scheduled for early 2025.

However, due to the performance of the in-house AI models not meeting expectations, the release was first postponed to 2026, with an internal plan to launch alongside iOS 26.4 in March. This plan remained unchanged until last month, as Apple executives did not want the new features to be delayed beyond spring 2026.

Recent internal testing has disrupted this plan. Testers reported that the new Siri has technical flaws, mainly related to response stability and accuracy. Issues include Siri failing to correctly handle queries, long response times, accuracy problems, and Siri interrupting users when they speak too quickly. More troubling is that the new Siri occasionally incorrectly calls upon integrated features with OpenAI’s ChatGPT instead of using Apple’s own technology, even when the request could be handled by Apple’s system.

As a result, Apple is adjusting its release strategy, planning to split the original set of new features into subsequent iOS versions. The most likely delayed features are two core functionalities: first, an expanded ability for Siri to access personal data, allowing users to search old messages, find podcasts shared by friends, and play them directly via voice commands; second, an advanced voice application control system that enables users to perform cross-application tasks with a single command, such as searching, editing images, and sending them to contacts. Although these features have initial support, their stability has not yet met Apple’s standards for release.

The repeated delays are likely related to Apple’s strict stance on privacy protection. During an employee meeting last week, software engineering chief Craig Federighi emphasized once again that personalized AI must not compromise user data. Federighi believes that the industry standard involves sending user data to servers for storage and training, but Apple aims to break this norm by leading the AI field: AI data will only reside locally or on privacy-protecting servers, with training relying on authorized information and synthetic data rather than real user data. He believes this approach can still deliver excellent AI experiences and will eventually be adopted industry-wide. Tim Cook also hinted during the same meeting that Apple is internally developing new data center chips, not only to enhance AI computing power but also to build a more “Apple-like” data center infrastructure tailored for its devices.

Meanwhile, Apple’s AI team experienced a significant talent drain last year, with dozens of core members leaving, including the head of the foundational models team and the Siri intelligent search project leader, who joined Meta. Several key researchers also moved to OpenAI, xAI, and Cohere.

Notably, Apple is still advancing two unrevealed new features, including a web search tool similar to Perplexity and a custom image generation feature based on Image Playground. These have appeared in the iOS 26.4 and 26.5 beta versions and may proceed partially as originally planned.

Apple is also developing a major AI project codenamed “Campo” for future systems like iOS 27. The plan is to transform Siri into a system-level assistant similar to a chatbot, deeply integrating advanced models like Google Gemini and the Apple ecosystem, covering core apps such as Mail and Calendar.

As early as August last year, reports indicated that Apple was in preliminary negotiations to use Google’s Gemini AI to improve Siri, with plans to pay around $1 billion annually for access to Google AI technology. In January, Apple reached a multi-year partnership with Google, with its next-generation foundational models based on Google Gemini and cloud technology. These models are expected to support future Apple smart features, including a major Siri upgrade scheduled for later this year.

However, system-level data access must balance user experience with privacy protection, so Apple may strengthen local processing and permission controls. Information security remains a long-standing priority for Apple.

Additionally, Apple is designing a new AI chatbot whose underlying models can be swapped out at any time, allowing the company to choose whether to use Google-powered systems in the future.