For privacy and security, think twice before granting AI access to your personal data

AI is increasingly integrated into nearly every aspect of our lives, from mobile phones and applications to search engines and even drive-through services. The emergence of web browsers with built-in AI assistants and chatbots signifies a notable shift in how individuals seek and consume information online compared to just a few years ago.

However, AI tools are frequently requesting extensive access to personal data, often under the pretense that such access is essential for their functionality. This level of data access is not only unusual but also should not become the accepted norm.

Not long ago, it was customary to question why a seemingly innocuous free application, such as a “flashlight” or “calculator,” would attempt to request access to your contacts, photos, or real-time location. These apps often did not require such data for their primary function but sought it to monetize user information. Today, the landscape with AI is strikingly similar.

Consider Perplexity’s latest AI-powered web browser, Comet, for instance. Comet aims to help users find answers using its integrated AI search engine and automate routine tasks like summarizing emails and calendar events.

During a recent hands-on review, TechCrunch observed that when Perplexity requested access to a user’s Google Calendar, the browser sought a broad array of permissions for the user’s Google Account. These permissions included the ability to manage drafts and send emails, download contacts, view and edit events across all calendars, and even copy an entire company’s employee directory.

While Perplexity states that much of this data is stored locally on your device, granting these permissions still gives the company rights to access and use your personal information, including for the purpose of improving its AI models for all users.

Perplexity is not unique in its requests for extensive data access. A growing number of AI applications promise to streamline tasks, such as transcribing calls or work meetings, but demand real-time access to private conversations, calendars, contacts, and more. Meta, for example, has also explored the boundaries of data access for its AI apps, including seeking permission to use Meta AI on photos stored in a user’s camera roll that have not yet been shared.

Signal President Meredith Whittaker recently characterized the use of AI agents and assistants as akin to “putting your brain in a jar.” Whittaker highlighted how some AI products offer to handle mundane tasks like reserving a restaurant table or booking concert tickets. However, to execute these tasks, the AI may require permissions to open your browser (potentially accessing stored passwords, bookmarks, and browsing history), utilize a credit card for payment, mark the event on your calendar, and even access your contacts to facilitate sharing the booking with friends.

There are significant security and privacy risks associated with relying on AI assistants that demand access to your data. By granting such access, you are instantly and irreversibly handing over a comprehensive snapshot of your most personal information, including years of inbox content, messages, and calendar entries. This extensive data transfer is typically justified by the promise of saving time or, as Whittaker suggests, reducing the need for active thought.

Furthermore, you are empowering the AI agent to act autonomously on your behalf, which requires an immense degree of trust in a technology known to be prone to errors, fabrications, or “hallucinations.” Relying on AI also necessitates trusting the profit-driven companies developing these products, which depend on your data to enhance their AI models’ performance. When issues arise, it is common practice for human personnel at AI companies to review private prompts to diagnose and resolve problems.

From a security and privacy perspective, a simple cost-benefit analysis reveals that connecting AI to your most personal data is generally not worth the trade-off. Any AI application requesting such profound levels of permissions should trigger immediate caution, much like a simple flashlight app demanding your real-time location. Given the vast quantities of data you relinquish to AI companies, it is crucial to critically assess whether the benefits truly outweigh the inherent risks to your privacy.