Token's digital assistant violated security rules by design

Token built Rehearsal, an AI agent that operates inside user browser sessions with full user permissions.

Because servers cannot distinguish human clicks from agent actions, the tool initially bypassed security rules by default, prompting Token to implement read-only limits and session recordings.

How the digital assistant works

The assistant, called Rehearsal, was designed to answer questions within Token's platform. Rather than simply giving instructions, it takes action on its own: opening browser tabs, setting filters, and filling out forms.

Because it runs within the user's session, it operates with the same permissions and access as the human being. Systems and computers cannot distinguish whether a click on the screen was made by the user or by the digital assistant.

The security measures implemented by the company

Because Token had control over both the assistant and the app, it was able to mitigate risks from within. Among other things, it implemented measures such as:

• Visual indicator on the screen that dims the application when the assistant is active.
• Recording of the movements of each session.
• Restrict the assistant to only see specific information, without being able to store data without the user's approval, fill out hidden or sensitive forms, or send unauthorized requests.

What does this mean for the future of the internet?

This protection was possible because Token built the app and the assistant at the same time. However, in practice, most services cannot know exactly who is performing an action.

↳

READ ALSO

HACKS.GR / RELATED

AI Unlocked

Judicial investigation into book fraud in Indian public libraries

↗

To address the problem, browsers like Chrome and Edge are introducing new features, such as consent requirements for automated browsing and per-site settings. At the same time, new standards are being developed to help apps and servers recognize when an automated assistant is visiting a page.

Until common solutions are established, security managers are called upon to identify programs and extensions that operate with artificial intelligence, record who uses them, verify user identities, and require explicit approval for any sensitive action.

>Stay connected with Hacks.gr

Set us as a preferred source on Google and find us on social media.

Preferred source

↗

end of article

Read the original English article on Hacks.gr