AI vendors can block specific prompt-injection techniques once they are discovered, but general safeguards are impossible ...