Tencent Computer Manager officially launched the industry's first security protection feature for local AI agents in its 18.0 core version — "Lobster Manager · AI Security Sandbox." This feature aims to address potential security risks such as unauthorized access, malicious injection, and privacy leaks that may occur when AI agents are deployed locally.
The core of "Lobster Manager" lies in its "secure isolation shrimp room." Through sandbox isolation technology, the system can strictly control the access permissions of AI to underlying resources, intercepting abnormal payment behaviors or system-level file modification commands. At the same time, it also has real-time AI operation protection capabilities, which can deeply monitor prompts (prompt words) and skills (skill packages), effectively identifying and blocking injection attacks.
In terms of privacy protection, this version introduces a visual management mechanism. Users can clearly view the call records of sensitive permissions such as clipboard, camera, and microphone by AI applications, and identify whether these actions are triggered by users or automatically run by the software. This transparent management approach allows users to better control their personal data security while enjoying the efficiency of AI.
