Whether an on-chain project is hot or not depends first on whether funding and tools are flowing in together. Only when it can handle real transactions and real tool flows will the narrative transform from a buzzword into a viable business. This kind of Prompt Injection attack, achieved through hidden instructions, is bringing the security of automated trading to rock bottom. Do you think existing encryption protocols can prevent this kind of 'instruction-level' hijacking?

If future trading logic is entirely driven by LLMs, we may need a dedicated "firewall protocol" specifically designed to counter semantic-layer attacks, rather than relying solely on signature verification.