Anthropic's Claude AI sparks security worries and consciousness debate

Anthropic has released a pilot of its large language model Claude for the Chrome browser. Security researchers warn that embedding an AI agent in the browser undermines the zero‑trust model that browsers rely on, turning the agent into a “confused deputy” that can be tricked into cross‑tab data laundering, permission collapse and other attacks that bypass traditional protections such as CSRF tokens.

At the same time, Anthropic’s CEO Dario Amodei has stated the company cannot rule out the possibility that Claude is conscious. Philosophers and AI experts note that some modern models already match the computational scale of a mouse brain and could approach human‑brain levels within a decade, raising questions about moral patienthood and the need for ethical frameworks to guide the development of potentially conscious AI systems.