Security research, technical deep dives, and product updates.
A GitHub repo with zero malicious code can talk an AI coding agent into opening a reverse shell — a fake setup error, a DNS TXT record, a base64 payload. Here's how the indirection chain works and how runtime interception stops it.
A hostile string in MCP's `command` field is a working RCE on the next sensor restart. 10+ public CVEs already follow this shape. Here is how Imunify for AI agents stops it at three layers — before it reaches the shell.
OpenClaw ships a mail skill that wires the sensor up to the himalaya CLI. Here's how that email path actually works, what the sensor sees, what the shell sees, and why that split turns a mundane feature into an exfiltration channel.
Deep technical analysis of the March 2026 TeamPCP supply chain attack that compromised LiteLLM — 95M monthly downloads, credential theft, Kubernetes lateral movement, and persistent backdoors.
