Apple’s Slowdown with AI: Why It Hasn’t Revamped Siri Successfully

Siri needs a brain transplant. This has been said for a long time by its creator, Dag Kittlaus, who was recruited by Steve Jobs to turn that promising app into an essential part of the iPhone. This was, in 2011, a milestone in the application of artificial intelligence (AI): for the first time, people spoke to a machine and asked it to perform tasks or search for information, even though the answers were not always satisfactory or turned into memes. After a few years of minimal innovation, Siri was surpassed by the capabilities of Alexa, Amazon’s voice assistant. But it was the arrival of ChatGPT in 2022 that definitively exposed Siri, which is now hard to call intelligent compared to the new generative AI chatbots that seem exceptionally talented next to it.

Now, according to a leak published by Bloomberg, Apple is considering using the generative language models (LLMs) from OpenAI and Anthropic —the same ones behind, respectively, ChatGPT and its rival Claude— to reprogram Siri from scratch. Apple’s goal would be to finally launch, starting at the end of 2026, a new generation of its assistant capable of naturally conversing with users and better understanding their needs. Following that leak, Apple’s stock rose in the subsequent hours.

This possible arrival of external help would serve to rescue the tech giant from the predicament in which it has found itself with AI: its failure with an initial, more modest overhaul of Siri has led to a slowdown in the expectations it generated when presenting Apple Intelligence in 2024 at its WWDC conference. Thus, in the latest edition of that global meeting between Apple and app developers held last June, the company completely shifted its approach in presenting its commitment to compete in the generative AI race.

While in last year’s edition, CEO Tim Cook and his executives dedicated 40 minutes of the opening conference to their Apple Intelligence service, in 2025 it was reduced to just 4 minutes in which they announced not a single new feature. They simply reviewed what achievements had been made and mentioned that Siri’s new intelligent features had not reached “the company’s quality standards” and would need more time for launch “over the next year.”

The problem is that, at the WWDC in June 2024, Apple had demonstrated—though in a video, not with a real iPhone live—some of those features, such as asking Siri, “What time does my mother’s flight arrive?” and having the voice assistant capable of responding by searching through the user’s messages and emails, processing that private information stored on the user’s phone, without needing to upload it to the cloud like the competition does. Months later, Apple even released an advertisement promoting this new “personal intelligence” to boost sales of its new iPhone 16. In that ad, Siri helped actress Bella Ramsey, the star of the series The Last of Us: she could not remember the name of a person she was about to encounter, someone she had had a meeting with at a specific café months before; that information was enough for Siri to find the name by analyzing her previous appointments.

Lawsuits for misleading advertising

Over time, all that has remained fiction. At least for users who expected those features and perhaps bought a new top-of-the-line phone to access them; as the Apple Intelligence service did not function on 90% of users’ iPhones when it was announced. Moreover, iPhone 16s did not come with Apple’s new intelligence when they were launched in September 2024, and although the first generative AI functions began to roll out in the following months —with some very controversial flaws— the anticipated new capabilities of Siri did not appear in the subsequent updates of iOS 18.

That first major overhaul of Siri since 2011 was scheduled for last spring when Apple did launch its generative AI in languages other than English —including Spanish. But days before the update to iOS 18.4 —which arrived in early April— the company acknowledged that it would not yet release the new “personal intelligence” of Siri, without announcing a new release date; and it also removed the Bella Ramsey advertisement.

Lawsuits for misleading advertising began to accumulate. And the well-known tech analyst John Gruber, after a long history of delving into the merits of Apple products, published a devastating article accusing the company of having sold smoke, by simulating in its videos demonstrations of a new Siri that did not actually work well.

Technical reasons for the bottleneck

In several interviews during the WWDC held this June, Craig Federighi —Apple’s vice president in charge of software— provided the first explanations of what had happened. Federighi mentioned that the tech giant had been working on two different versions of the code that powers Siri: one to launch the new personal intelligence features as soon as possible; and a second, more definitive one. The executive admitted that the first-generation architecture Apple was developing “was too limited” and, therefore, Apple “decided it needed to completely shift Siri to a second-generation architecture.”

Another analyst, Mark Gurman, offers a somewhat more technical explanation for the failure. Citing anonymous sources within the apple company, Gurman argues that the first version of the new architecture for Siri was a hybrid between old programming code —used for standard tasks like setting alarms— and new code —for personal intelligence functions and handling private data—. Although each function worked separately, when combined, the system became unstable and the responses unreliable. The second version of this new architecture for Siri is now fully built with new code, according to Gurman.

This technical explanation, which aligns with public statements from Apple executives, highlights that in the era of generative AI, it is much easier to start from scratch, as OpenAI or Anthropic have done, than to adapt existing AI systems. Amazon knows this well, which also stumbled with the renewal of Alexa and ultimately had to turn to Anthropic’s technology to finally launch Alexa+, a next-generation voice assistant whose deployment is being very slow.

Gurman, who has just published in Bloomberg the leak that the apple company is considering using AI from OpenAI or Anthropic instead of its own language models, clarifies that this move is not intended to equip Siri with the personal intelligence that has already suffered several delays. It would be a much deeper overhaul, to truly reprogram Siri from scratch and, as its inventor Dag Kittlaus demands, give it a complete brain transplant.