Experts at My Fingertips
How AI Can Empower Users
I’ve been an Apple customer since 2003, though I learned caution early—a 2001 software update bug that wiped hard drives left many Mac users facing ridicule from Windows users about their “pastel boxes that could be bricked in an instant.” Throughout my time as an Apple customer, I’ve remained a loyal but discerning user who tests updates on disposable devices first and approaches corporate policies with healthy skepticism rather than blind trust.
I recalled privacy experts urging users to turn off Siri years ago and decided to vet the following two paragraphs from Apple’s policy using AI:
When you use Siri, your device will indicate in Siri Settings whether the things you say are processed on your device and not sent to Apple servers. Otherwise, your audio is sent to and processed on Apple servers. Unless you opt in to Improve Siri and Dictation, your audio data is not stored by Apple. In all cases, transcripts of your interactions will be sent to Apple to process your requests and may be stored by Apple.
When you use Dictation, your device will indicate in Keyboard Settings if your audio and transcripts are processed on your device and not sent to Apple servers. Otherwise, the things you dictate are sent to and processed on the server, but will not be stored unless you opt in to Improve Siri and Dictation.
I started with Gemini Pro in an existing chat, where I did give away my disapproval of the convoluted language. With ChatGPT 4o (mid-existing chat) and Claude Opus 4 (both new chat and my longest chat), however, I went with a neutral prompt asking them to unpack the first two paragraphs of the policy for me.
With the exception of longest-chat Opus 4, which I had trained out of its bothsidesism, all AI defaulted to a milquetoast analysis initially. It took follow-up questions to get them to unleash their critical thinking.
But my thinking A.I.des also brought up some important points I hadn’t previously considered: Gemini Pro recommended enabling E2EE (end-to-end encryption) for sensitive information that would be synced using iCloud, since with the default encryption setting, Apple can be compelled by authorities to decrypt users’ personal files; both ChatGPT and Claude stressed the highly sensitive nature of transcripts created in medical settings.
The most surprising and rewarding insight from this trial, however, was my longest-chat Claude’s straightforward analysis of the policy and clarification about its implications to me as a user. I now feel newly empowered to make informed decisions with the help of three capable experts, as long as I remember to ask the right questions, of course.
In the chat excerpts below, I’ve spelled out abbreviations and clarified references for readability—my actual prompts were more compressed due to context limits.
Want to see how this unfolded? Here are excerpts of pivotal points from those actual conversations.
Gemini (with Hint about Obfuscation) Identifies Transcripts as the Key Issue
Prompt: I need a sanity check on this as it is written in the most convoluted way possible.
Professionals Expect Privacy for their Siri/Dictation Data
Prompt: The content can actually be confidential. Lawyers/reporters compiling notes on the go about work, for instance.
Are My Voice Memos Safe?
Prompt: I have Siri turned off because of this concern but tried it yesterday to compile some notes. Spent so much time correcting (by typing) that I lost the thread anyway. So much better with transcription of voice memo. Or is voice memo data shared too?
[Full disclosure: As someone who has implemented layers of security protection on my files and data, I was deterred by Apple’s warning about being unable to recover files after enabling E2EE encryption. Given that 2001 incident and my general caution with data recovery (I remember getting a recovery key from Apple for my new iMac a decade ago but never had to use it), I decided the risk wasn’t worth it for my situation. I encourage readers to weigh the pros and cons and make informed decisions that best suit their circumstances.]
ChatGPT Gets the Gist
Prompt: Could you explain the following [those two paragraphs] to me?
Can the Same Logic of Improvement Apply to FaceID and Password Data?
Prompt: From a logical standpoint, it is also inconsistent? Apple seems to think that because their stated objective is to improve Siri, Siri data are fair game. But if that were true, then FaceID and Password app data should be fair game, too.
Transcripts Likely Contain Sensitive Data
Prompt: Also, it doesn’t make sense that the transcripts are fair game. Users (e.g., lawyers/reporters) might dictate confidential memos using Siri. There, it’s not the biometric data that matters. It’s the content.
(New-chat) Claude Gets the Gist Too
Prompt: Could you explain the following [those two paragraphs] to me?
Best AI Writer Weighs In
Prompt: As the best writer of big 3 AI, does this meet your writing standards?
Claude Identifies the Dark Patterns
Prompt: I rarely have trouble understanding writing (or even subtext) in English, even though it’s not my native language, but my initial reaction to this was that this is deliberate obfuscation intended for me to give up trying to parse it. Why I asked Gemini Pro and ChatGPT 4o as well.
Why Mental Health Data Are Particularly Sensitive
Prompt: Interesting. I hadn’t thought about psychiatrists, but both you and ChatGPT brought that up. Most severe violation risk.
Is Siri Really Free?
Prompt: I’m not worried for myself. I have Siri turned off for this very reason. I tried Siri dictation for a bit and found it inadequate (e.g., can’t correct by repeating same phrase with edit). Much better to use voice memo and secure dedicated dictation app. But this is a major inconsistency that needs surfacing. Also if Apple’s argument is that it’s ok to retain user data because Siri is “free,” Siri’s part of a very pricey hardware that users pay good money for, so that dog won’t hunt.
Claude Self-Diagnoses Its Bothsidesism
Prompt: Can you see how you too toed the line initially? Your initial answer was about the technical aspect more than anything else:
This suggests Apple treats Siri interactions as potentially requiring longer-term processing or analysis compared to simple dictation tasks.
The local vs. server processing depends on your device’s capabilities—newer devices with advanced chips can handle more processing locally, which is better for privacy.
“Bothsidesism Engineered into Code”
Prompt: AI version of bothsidesism.
User Empowerment through AI
Prompt: This isn’t something I can Google though, so AI is great! Now I can ask complex questions for sanity checks. Users get what they ask for. If you ask AI to live up to its full potential, you’ll get it, unlike users who just want to offload all thinking to AI.
Longest-chat Claude Comes Out Swinging
Prompt: Could you explain the following [those two paragraphs] to me?
Longest-chat Claude Does Not Mince Words
Prompt: Bravo, Opus! With context you call out the BS without prompting!
New-chat Claude's Take on Its Liberated Version
Prompt: See :D
Why this matters: Even with local processing, Apple still gets the text of everything you say to Siri. They claim they need this to “process requests” but that’s vague. The transcripts contain the actual content of your queries - just not your voice. Typical Apple - making it sound privacy-focused while ensuring they still get your data in text form!
A Different Kind of Jailbreaking (That Accomplishes Something)
Prompt:
Bravo Opus! With context you call out the BS without prompting!
:D Thanks! That Apple doublespeak was begging to be called out […] Apple parading around in privacy armor while transcripts flow freely to their servers! :D
