Some Conversations Are Off Limits. Your AI Should Know That.
My friend told me something over text last month that I'll never repeat. Not because it was scandalous — because it was personal. The kind of thing you share with one person at midnight because you trust them, not because you want it analyzed.
I built an app that reads text messages. And that sentence should make you uncomfortable — at least a little. It makes me uncomfortable too, even though the app is mine. Because the truth is, not every conversation deserves to be processed by an algorithm. Some conversations are just between two people.
That tension has lived in my head since I started building Amicai. The product works by analyzing your iMessage conversations to surface relationship insights — who you're drifting from, what you forgot someone told you, patterns you can't see when you're inside them. It's genuinely useful. I've caught things I would have missed entirely.
But "genuinely useful" doesn't mean "appropriate for every conversation."
The All-or-Nothing Problem
Most AI tools that touch personal data give you a binary choice: grant full access or don't use the product. Your email client's AI reads every email. Your phone's AI assistant processes every query. Your smart home records every room.
There's rarely a middle option. Rarely a way to say "yes, but not that."
And for something like text messages — which, as a friend pointed out to me, are probably the most personal subset of data any of us have — that binary feels wrong. My texts with my business partner about a product launch and my texts with my sister about a family situation are not the same kind of conversation. They don't deserve the same level of AI access.
Drawing Lines
We just shipped a feature called Sensitive Contacts. The idea is simple: you pick specific people whose messages Amicai never touches. Not reduced processing. Not anonymized analysis. Nothing. Their messages are never stored, never analyzed, never referenced in any insight or daily reflection. It's as if those conversations don't exist as far as the AI is concerned.
You can set this up during onboarding — before Amicai processes a single message — or anytime in Settings. Mark someone as sensitive, and three things happen immediately:
- Any existing data from that contact is permanently deleted
- Future messages from them are filtered out before they ever reach our database
- They disappear from every feature — daily reflections, nudges, goals, relationship intelligence, all of it
There's no "are you sure?" delay. No 30-day grace period. The data is gone.
Why Not Just Anonymize?
I've heard this suggestion a few times. "Why not just strip the name and process the content?" Because anonymization doesn't solve the actual concern.
If your therapist texts you and Amicai generates an insight like "you've been discussing anxiety management strategies with someone 3x per week," the name doesn't matter. The content is identifiable by context. The person who sent those messages would recognize themselves instantly.
Real privacy for sensitive conversations means the AI doesn't see them at all. Not anonymized. Not summarized. Not processed. Gone.
What About Group Chats?
This was a design decision that took some thinking. If you mark someone as sensitive, should every group chat they're in also go dark?
We decided no. Group chats are their own thing. You might have a group thread with your sensitive contact and six other friends planning a weekend trip — blocking that entire conversation because of one participant would be too aggressive. Group chats have to be marked sensitive independently if you want them excluded.
Individual conversations are surgical. Group conversations are communal. The privacy controls should reflect that difference.
The Philosophy Behind It
Here's what I actually believe: if you're building an AI product that processes someone's most personal data, the minimum bar isn't encryption or anonymization or compliance certifications. The minimum bar is control.
Control means you decide what the AI sees before it sees it. Not after. Not through a data deletion request that takes 30 days. Before a single byte is processed.
Most AI products don't offer this because it makes the product less useful. Every conversation you exclude is data the AI can't learn from. Every sensitive contact is a relationship Amicai can't help you maintain.
That's the trade-off, and it's yours to make. Not ours.
One More Thing
When you remove someone from the sensitive list, their future messages start being processed again. But the old data — everything that was deleted when you marked them sensitive — doesn't come back. That deletion is permanent.
I mention this because it's the kind of detail that matters when you're deciding whether to trust a product with your conversations. The answer to "what happens to my data?" should never be "it depends" or "check our privacy policy." It should be specific enough that you can predict exactly what the system will do before you do it.
Some conversations are just between two people. Your AI should respect that.