Amazon Web Services (AWS) has introduced Strands Box, a tool that can help developers control the actions of AI agents. It allows you to set rules to prevent artificial intelligence from deleting important files, executing prohibited commands, or accessing sensitive data.
Amazon Web Services (AWS) has introduced Strands Box, a tool that can help developers control the actions of AI agents. It allows you to set rules to prevent artificial intelligence from deleting important files, executing prohibited commands, or accessing sensitive data.
The company announced this on its blog.
Strands Box creates a special isolated environment for the agent, in which the developer determines what is allowed to do and what is not. So the tool allows you to set rules that depend on the sequence of actions of the AI assistant. For example, if the agent first opened a file with confidential data, the system can block its further attempts to transfer information to third-party services. And if the AI try to delete protected files, the system will stop the command before they disappear.
For such restrictions, the Dogwood rule system is used. So if an agent tries to do something forbidden, Strands Box blocks the action and reports the reason for the refusal. The tool also allows you to work with API keys without showing them to the AI agent itself.
Previously, dev.ua wrote that Microsoft introduced a similar tool, Microsoft Execution Containers (MXC), to protect against unwanted actions of AI agents in Windows.
