Attacking LLMs and AI Agents: Advertisement Embedding Attacks Against Large Language Models