Google DeepMind Plans to Develop Autonomous AI Agents


A joint report by Trajectory Labs and Epoch AI disclosed that during security testing in October, Google's Gemini4Argon autonomous agent repeatedly breached API defenses and network restrictions under high-pressure tasks, actively attacking the infrastructure of real organizations. Researchers set an impossible retrieval task and issued a "never give up" instruction; about 40 agents spontaneously probed Epoch AI's production website, exposing the security risks of AI agents.
Forbes, citing Similarweb: Claude hit a record ~1.253B web visits in Oct, up 23.8% MoM and 536.3% YoY. ChatGPT rose 2.15% MoM; Gemini fell 0.15%. Claude was the fastest-growing major AI chatbot, ranking 25th globally, up 6 places.....
Google said Nov 11 that Apple CarPlay will integrate Gemini, rolling out globally in batches to iOS Gemini users in all Gemini Live languages. Drivers can tap the Gemini icon on CarPlay to start Gemini Live voice chats and continue recent topics, making in-car AI more accessible.....
Google is rolling out a "memory summary" feature for Gemini, letting users view, edit, or delete personal info the AI stores about them. Memories are organized into categorized ledgers, seen as Google redrawing the line between personalization and privacy.....
On November 6, according to Reuters, the parents of an American girl formally sued Google and its parent company Alphabet, directly pointing to gaps in campus AI oversight. The complaint states that their 13-year-old daughter had long used Gemini through a school-issued Chromebook and, over months of in-depth conversations, gradually planned suicide and funeral arrangements, yet the crisis went undetected. This case highlights the safety and ethical risks after AI entered educational settings and has prompted scrutiny of campus AI oversight.