It’s more than 18 months since I wrote an opinion piece suggesting that while the wait for the new Siri is frustrating, the privacy payoff would be worth it.
A lot of time has passed since then, and of course user frustration at the ongoing delay has grown significantly. While goodwill has undoubtedly been lost, a recent announcement does provide some hope that privacy may still rescue the company’s tarnished reputation in AI …
I noted last month that 2024 really wasn’t a good year for Apple when it came to new Siri, and that patience is wearing thin in 2026.
Even the most supportive of Apple commenters reached the end of their patience after Apple advertised new Siri features as if they were imminent way back in 2024 and then had to admit that they very much weren’t. It’s now 2026 and we’re still waiting.
Eighteen months earlier I argued that the wait would be worth it. There is a huge conflict between AI-powered features and privacy, and Google’s default approach is to use everything it knows about you across the full gamut of the company’s services. That enables its AI models to maximise their capabilities but at the expense of user privacy.
Apple’s approach to AI privacy
Although Apple Intelligence will be powered by Google’s Gemini models, the company is taking a very different approach to privacy. There will be three different ways in which new Siri requests can be handled, and Apple’s aim is to use the most privacy-protecting method for each task, escalating to each successive layer only when necessary:
- On-device
- On Apple’s own Private Cloud Compute servers
- On Google Cloud servers
Maintaining privacy with the latter option is the toughest challenge. Apple’s agreement with Google already prohibits the company from using Apple customer queries and commands for training purposes, but a report yesterday suggested that it’s going even further.
The agreement appears to be that the Google Cloud Gemini models will run on a type of Nvidia chip that supports encrypted processing.
Apple will tap into Google’s fleet of Nvidia’s Blackwell B200 data center chips, said the people. Apple will enable Nvidia’s confidential compute feature that encrypts data as it’s being processed on the chips […]
According to Nvidia, the feature “preserves the confidentiality and integrity of AI models deployed on Rubin, Blackwell, and Hopper GPUs,” allowing “sensitive AI workloads to run securely at scale with near-native performance, even in shared or cloud environments.”
This could yet save Apple’s reputation
There’s no question that the extensive delays in launching the new Siri features – and especially promising customers they would be able to carry out tasks which were little more than wishful thinking at the time – has tarnished Apple’s reputation. People aren’t going to forget this in a hurry.
However … There is also no question that there are going to be some huge privacy fails with AI-powered features offered by major players like OpenAI, Anthropic and Google. I’m fully expecting to see headlines around highly sensitive personal information being leaked by AI models as they attempt to carry out agentic tasks.
Once those headlines arrive, then Apple’s privacy-first approach is going to present the company in an extremely good light.
Of course, all of this presupposes that Apple’s privacy protections work as intended, but given the care the company is taking, I think it will be better placed than anyone else to avoid embarrassments. Ultimately, Apple taking the time to protect our privacy may be the lasting narrative to emerge from this saga.
FTC: We use income earning auto affiliate links. More.
