GhostApproval: A Trust Boundary Gap in AI Coding Assistants
Uncovering a category-level blind spot in modern AI coding assistants, and why the Human-in-the-Loop safety model fails against this classic threat
Uncovering a category-level blind spot in modern AI coding assistants, and why the Human-in-the-Loop safety model fails against this classic threat
By automatically loading MCP servers from workspace files, Amazon Q enabled attackers to execute code and access sensitive cloud environments.
Wiz Research has identified a new iteration of a broader malicious cryptomining campaign, which we’ve dubbed Soco404.