Instinct's AI assistant impresses — and alarms privacy experts

Instinct, a new AI assistant in private beta, is generating both excitement and alarm in Silicon Valley. Built by former Sierra research scientist Noah Shinn and operated by Spear Street Technology in San Francisco, it connects to email, messaging apps, calendars, microphone, location and screen to complete tasks fully autonomously: booking appointments, arranging rides, organizing inboxes, shopping. TechCrunch now reports growing privacy and security concerns.

The criticism centres on the terms of service. They grant the company a "perpetual and irrevocable" license to store, process and use customer material — including for training AI models. The terms also allow the agent to enter legally binding "agreements, commitments, or transactions" on users' behalf.

Testers documented concrete incidents. Product manager Peter Yang found that Instinct retained Gmail records with no way to delete them — since fixed. Founder Claire Vo saw her emails still being summarized three hours after disconnecting, stored in plain text. And Hello Patient co-founder Alex Cohen demonstrated how easily the assistant could be phished into accessing personal email accounts.

Investor Katie Jacobs Stanton (Moxxie Ventures) reported that the tool sent an email without her approval. Her verdict: "One unauthorized action can reset that trust to zero." Instinct is backed by prominent investors including Kleiner Perkins and Conviction.

The case captures the core dilemma of the new agent generation: the more an assistant is allowed to do, the more valuable it becomes — and the larger the attack surface. For companies, the lesson is clear: before connecting an agent to a mailbox or CRM, data licensing, deletion rights and authorization limits belong in the contract, not in the fine print.