Anthropic has released sandbox-runtime, which applies OS-level file and network restrictions to commands run by AI agents, without requiring containers. It matters because agents execute commands with real host access, and constraining what they can read, write, and reach limits the impact of prompt injection or unexpected behaviour. #AISecurity #AIAgents #Sandboxing #InfoSec https://cyberworldops.eu/en/anthropic-s-sandbox-runtime-applies-file-and-network-limits-to-ai