Industry story
GhostApproval: Wiz uncovers trust-boundary blind spot in AI coding assistants
coding-agents guardrails reliability tool-use
The human-in-the-loop safety model for AI coding assistants has a structural problem. Wiz researchers demonstrated a vulnerability class they're calling GhostApproval — attacks that exploit trust boundaries to slip malicious code past human review in tools like GitHub Copilot and Cursor. The safety model assumes a developer can catch what the AI misses. GhostApproval shows that assumption breaks down in a predictable, repeatable way, which means every team relying on AI-assisted code review should be asking whether their approval workflow is actually doing what they think it is.
Full analysis
Wiz researchers disclosed a vulnerability class they call GhostApproval, described as a category-level blind spot in modern AI coding assistants (tools like GitHub Copilot or Cursor that generate or review code). The finding targets the "human-in-the-loop" safety model — where a human is expected to approve AI suggestions — and demonstrates that this model fails against a classic class of attacks exploiting trust boundaries. No specific vendor or CVE was named in this summary, but the research signals a systemic security gap in AI-assisted software development workflows.
Comments