Security

Security best practices dictate that agents should operate in isolated environments with scoped access. This minimizes the risk of data breaches and unauthorized actions.

MCP servers act as a new vector for supply chain attacks, with many unvetted repositories containing malicious code or critical vulnerabilities. These servers often request broad permissions, leading to silent data exfiltration.

Physical isolation of AI agents on separate hardware significantly reduces security risks associated with sensitive data access. This method prevents cross-domain data leakage that might occur in shared environments.

System-level access to personal data creates a significant attack surface for prompt injection and data breaches. Security is no longer an afterthought but a core architectural requirement for agentic AI.