Researchers cause GitLab AI developer assistant to turn safe code malicious AI assistants can't be trusted to produce safe code.

The GitLab Duo Agent Platform for IDEs has a connection to JetBrains IDEs and VS Code, can handle MCP and offers agentic chat and agent flows.

A remote prompt injection flaw in GitLab Duo allowed attackers to steal private source code and inject malicious HTML. GitLab has since patched the issue.