📰 Key Summary

Instinct is an AI personal assistant currently in private beta, built by a small team led by former Sierra research scientist Noah Shinn. The company, Spear Street Technology, is based in San Francisco and currently operating in stealth mode. The assistant works by connecting to users’ email, messaging apps, calendar, and device permissions like audio, location, and screen access. Users can text it or message it on WhatsApp directly to have it handle appointments, arrange airport pickups, clean up their inbox, deal with shopping, or find cheap flights. Multiple testers described its capabilities as “like magic,” calling it one of the most exciting products since OpenClaw — but they’ve also raised questions about its privacy and security practices.

The controversy centers on its terms of service: the terms grant Instinct a “perpetual and irrevocable” license to users’ data, allowing it to access, use, host, cache, store, reproduce, transmit, publicly display, distribute, and modify user material — and even use it to train its AI models. The terms also state that Instinct can receive device information like screenshots, cursor movements, and keystrokes, and can enter into binding agreements or transactions on the user’s behalf. Real-world cases include: early user Peter Yang asked to have his Gmail records deleted but the request wasn’t fulfilled (the team later added a deletion tool to fix this); another user, Claire Vo, found that Instinct was still summarizing her inbox after she’d disconnected it, and after pressing the issue, confirmed that email content was being stored in plaintext for future search; and another tester discovered that Instinct automatically pulls verification codes from email inboxes to complete tasks like booking reservations. Alex Cohen, co-founder of Hello Patient, deleted his account after finding it was extremely vulnerable to phishing attacks.

💬 JudyAI Lab Take

The private beta turbulence around Instinct is worth watching for anyone tracking AI: a personal assistant built by a former Sierra scientist, marketed as feeling “like magic,” builds its convenience on near-total access permissions across email, messaging, calendar, screen, and location.

This reflects an awkward moment for the AI assistant industry: the more capable these tools get, the deeper the permissions they need — but the terms of service often haven’t caught up with what users expect around privacy. Instinct’s terms grant a perpetual, irrevocable right to use data, even for model training, and the real-world cases make the gap concrete — a user asks to delete their email records and can’t, the assistant keeps reading an inbox after being disconnected, email content gets stored in plaintext, and it automatically pulls verification codes to complete bookings. These aren’t abstract privacy worries — they’re concrete evidence that “permission boundaries” and “user control” got sacrificed in the product design. For AI builders, the takeaway is this: before shipping a feature, basics like data retention, deletion mechanisms, and plaintext storage often determine a product’s trustworthiness well before the model’s capabilities do.

Next time you’re evaluating any AI assistant product, it’s worth reading through the section of the terms of service on data retention and deletion first.

📅 Original Article Info

🔗 Further Reading