[key points]

Key points

• •The article analyzes the risks associated with personal assistants based on language models (LLMs), particularly with the emergence of OpenClaw, a tool for creating customized assistants.
• •OpenClaw, while offering great flexibility, exposes users to significant security vulnerabilities, including hijacking via prompt injection.
• •The article explores the challenges of risk management and the implementation of protections against prompt injection attacks, highlighting the need to find solutions to secure AI personal assistants.