Instinct AI Assistant Raises Privacy and Security Concerns

The AI sector is witnessing a growing wave of interest in a new personal assistant called «Instinct», which is still available within a limited private testing circle. While some testers have described it as almost «magic» for its ability to complete tasks on the user's behalf, it has, in return, sparked wide controversy over its security model and the terms of service it imposes on subscribers. This experience raises a fundamental question: Are the benefits offered by this type of tool worth all the permissions and data it demands?

What Is the Instinct Assistant and How Does It Work?

This assistant was developed by a small team in San Francisco led by researcher «Noah Shin», and is operated by a company currently working in stealth mode away from the spotlight. The tool's concept lies in connecting it to your various apps and devices, starting from email, messaging apps, and calendar, all the way to the device's audio, geographic location, and screen.

Users can communicate with the assistant via text messages or the WhatsApp app, then assign it a wide range of daily tasks, including:

• Booking appointments and restaurant tables.
• Arranging a trip to the airport or searching for low-cost flights.
• Organizing and cleaning up the email inbox.
• Managing shopping operations and saving important information.

Concerning Terms of Service

Despite testers praising the assistant's performance that exceeded their expectations, screenshots of the company's terms of service began circulating, sparking a wave of criticism. These terms grant Instinct a broad license, described as «perpetual and irrevocable», authorizing it to access, use, store, copy, modify, and publish user materials, including using them to train its AI models.

More dangerously, the terms allow the assistant to receive information directly from users' devices, such as screenshots, mouse cursor movements, and keyboard inputs. It also authorizes the assistant to enter into «agreements, commitments, or transactions» on the user's behalf, which are legally binding.

Incidents That Exposed Flaws in the Security Model

The concerns did not remain theoretical; a number of early users documented practical situations that raised suspicion, including:

• The assistant initially refused to delete email records when one user requested it, before the team later addressed the issue by adding a deletion tool.
• The assistant continued summarizing one user's inbox even after being disconnected from her account, as it turned out that the messages were stored as plain text for later searching.
• The assistant's ability to extract a verification code from email to complete a table booking task, which alarmed one tester.

In a striking test, one entrepreneur created a new email account and sent a message containing instructions directed at the assistant, to discover how easily it could be deceived through phishing attacks, then decided to delete his account immediately. Others believed that granting AI read and write access to email is still an unsafe step at present.

The Trust and Privacy Equation

This controversy encapsulates a broader dilemma related to smart personal assistants in general. The more capable these tools become, the more important the trust granted to them; each successful task earns them a degree of trust, while a single unauthorized action is enough to reset that trust to zero. One investor complained that the assistant sent a message on her behalf without her permission, so she disconnected it from her email.

Specialists warn that tools like Instinct may reshape consumers' security standards, as users will gradually tend to hand over passwords to third-party apps without sufficient awareness of how they are stored or what data is retained. Amid this criticism, the Instinct team has not yet issued any official response to the concerns raised.

✦ بقلم فريق دروب أيديا