AI SecurityThe software wrapper around your AI agent is the real security risk
Researchers broke into official AI automation tools from Anthropic, Google, and OpenAI, not by tricking the AI itself, but by exploiting the ordinary code that connects it to the real world.