Instinct’s powerful AI assistant is raising privacy and security concerns
Instinct, a private-access AI personal assistant created by former Sierra researcher Noah Shinn's company Spear Street Technology, has drawn both acclaim and alarm from early testers. While praised for its capabilities in handling tasks such as booking appointments, managing emails and finding cheap flights, testers have raised concerns about its security model and terms of service, which grant the company extensive rights over users' personal data and device information.
The assistant connects to users' email, messaging apps, calendar, and device audio, location and screen, and can be controlled via text or WhatsApp. Its terms of service reportedly grant a "perpetual and irrevocable" licence to access, store and use user materials, including for AI training, and permit collection of screen captures, keystrokes and cursor movements, as well as allowing the AI to enter binding agreements on users' behalf. One early adopter, Peter Yang, found the company would not delete his Gmail records on request, though a deletion tool was later added after the issue was flagged.
- Instinct's AI assistant impresses testers but raises major privacy fears
- Terms of service grant broad, perpetual rights over user data
- One user found Instinct initially refused to delete his stored Gmail data
New here? Start with this
Instinct is an AI personal assistant built by Spear Street Technology, a company founded by Noah Shinn, who previously worked at the AI firm Sierra. Unlike widely available chatbots, access to Instinct is currently limited to early testers, who have been trying it out for everyday tasks such as booking appointments, sorting emails and finding cheap flights.
To do these tasks, Instinct needs deep access to a user's digital life. It can connect to email, messaging apps and calendars, and can also tap into a device's audio, location and screen, with users able to control it through text messages or WhatsApp. This level of access is what sets it apart from more limited assistants, but it is also the source of the concerns now being raised.
Those concerns centre on how much control the company has over user data once it is granted access. Reported terms of service allow the company to keep and use material indefinitely, including for training its AI systems, and to collect detailed activity such as keystrokes and screen captures. This matters because it raises broader questions, relevant to AI assistants generally, about how much personal data users should hand over in exchange for convenience, and how that data is protected once it is collected.
Both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
Advocates of Instinct's approach argue that a genuinely useful AI assistant cannot perform tasks like booking flights, managing emails and drafting agreements without deep, persistent access to a user's digital life, and that broad permissions are the necessary price of powerful automation. They contend that early-stage products often carry expansive terms of service to allow rapid iteration and model improvement, and that sophisticated users who volunteer for private-access testing are making an informed trade-off between convenience and exposure. Proponents also note that the company responded to criticism by adding a deletion tool, suggesting a willingness to adjust practices as the product matures rather than a fixed disregard for user control.
The case against
Critics argue that granting a company perpetual, irrevocable rights to personal emails, keystrokes, screen activity and even authority to enter binding agreements on a user's behalf goes far beyond what is reasonably necessary, however useful the assistant. They point to the initial refusal to delete a user's Gmail data on request as evidence that data rights promised on paper are not honoured in practice, undermining trust in the company's stewardship of highly sensitive information. For these critics, the values at stake are informed consent and proportionality: terms so broad and technical that few users can meaningfully evaluate them, covering capabilities as intrusive as audio and screen capture, represent a structural imbalance of power between company and user that convenience alone cannot justify.