Anthropic is scrambling to contain the leak, but the AI coding agent is spreading far and wide and being picked apart.
Four vulnerabilities in CrewAI could be chained together via prompt injection for sandbox escape, remote code execution, and ...