Using closed AI models from providers like Anthropic and OpenAI can create a fiduciary risk for enterprises by exposing company IP