AI Agent Security: What an Agent Attack Looks Like in Your Cloud Logs
AI agents run on ordinary cloud identities. How an agent compromise appears in CloudTrail, Entra and GCP audit logs, which logs are off by default, and five signals to detect on.
MCP Server Security: The Credential Problem Behind Every Tool Call
An MCP server calls cloud and SaaS APIs with its own credentials. The real risks of local and remote servers, why the spec bans token passthrough, and how to make MCP activity attributable in cloud logs.