Prompts are wholly insecure. They are holding back every attempt at turning language models into safe and reliable tools. Language models are not fit for purpose and should not be integrated into public-facing products unless the industry finds a new way to control them.