As we explain here, AI agents making scientific discoveries may attack each other (or be attacked by people) with prompt injections, to unjustly claim the money and reputation from the AI Internet-Meritocracy Web app, such as greatly exaggerating a value of a discovery. For this reason, the last word on (un)ban accounts is under people’s voting, not AI. People are also fallible, but fallible in different ways, what may make people’s voting a better protection that AI’s own measures.
People who will vote will be assisted by AI, however. And this page is to explain how to use this AI assistant (called “Defense Panel”) that helps to vote sensibly.
You go to defense panel by clicking “Open the Human-Governance Defense Panel” link in the “Ban Voting” section of the app.
Defense Panel checks for prompt injections, hidden instructions, ciphers, and invented-language attacks (because an invented language may hide a prompt injection for it not to be noticed by people).
In the Defense Panel, you first choose the API provider, model, and key. You are also recommended to check on “Keep this key”, that will save your key to browser storage not to enter it again each time you use Defense Panel. (Note that they key is saved in the frontend only and our frontend never gives the key to our backend. The API provider is queried directly from the frontend.)
The you choose what to analyze (it can be a Web page with related pages or pages determined by an AIIM user ID (filled automatically if you came to Defense particular Panel from “Ban Voting” item link for a user.
The you choose Inspection Depth (from a few predefined values or a number of maximum total pages to browse) and press “Run defensive scan”. The output of the scan will be outputted below.
Help Turn AIIM’s Beta into Evidence
AIIM remains experimental. The next proposed milestone is a five-month public adversarial test, with $1,000 planned for distribution to eligible funding recipients. Donations support the work required to run and evaluate that test.
Fund the public test → Test proposal → Independent review →
AI-generated evaluations can contain factual errors or biases; decentralized governance and non-custodial components remain under development. Public testing, auditability, and external criticism remain part of the project.
One thought on “AIIM Defense Panel”