What AIERC-193 is
AIERC-193 is a working mechanism — built and running today — that puts an AI safety check in the path of a token transaction. Before value settles, the transaction is evaluated and given one of two verdicts:
CLEAN the transaction proceeds. FLAGGED the facts are surfaced and the transaction is held or stopped, depending on the chain.
It is deliberately binary — CLEAN or FLAGGED. There are no risk scores, no tiers, no percentages shown to the user. A transaction is either fine to proceed or it is something you should see before it goes through.
The one idea to hold onto: the AI does the thinking, and that thinking happens off the chain — there is no AI model running inside a blockchain. What changes from one chain to another is where that AI verdict has its effect, and whether it leaves a record on the chain itself. That distinction is the whole document.
Where the AI lives
The AI's computation always runs off-chain (no language model executes inside a virtual machine). But "is the AI on-chain?" is not a question about where the computers sit — it is about effect and record. By that measure there are two very different situations:
On other blockchains (over RPC)
RPC = the connection a wallet uses to talk to a blockchain.
The AI is fully off-chain. It can only act before a transaction is broadcast — through the AIERC connection layer and the browser extension. It has no effect inside the chain and leaves no record on it. It protects the user; it does not become part of that chain.
On Chaingentic (our chain)
Chaingentic is the AIERC Layer-2 (L2) blockchain.
The AI is on-chain. It pauses and clears each transaction as part of producing the block, and it writes a record of what it did onto the chain's block explorer. The AI is a participant in the chain itself.
This is the point of having our own chain. Everywhere else, AI protection is an off-chain overlay riding on top. On Chaingentic, the AI graduates into the chain — gating transactions and leaving an on-chain record. On-chain AI is exclusive to Chaingentic.
On Chaingentic: when the transaction is checked
This is the full mechanism, and the precise answer to "when does a transaction get checked?"
- A transaction is submitted. As the sequencer (the part of the chain that assembles the next block) goes to include it, it places that transaction into an AI-pending lane — and pauses that one transaction, before the block is sealed.
- Only that transaction pauses. The chain never pauses. Every other transaction keeps flowing and blocks keep being produced. We pause the transaction that needs a look — never the blockchain.
- The off-chain Consensus Council — an independent panel of multiple AI models — evaluates the transaction and returns a single verdict.
- CLEAN the transaction is admitted into a block. FLAGGED it is dropped and never included.
- The AI's activity for that transaction is written as a memo on the Chaingentic block explorer — a plain record of what the AI did, anchored to the transaction. The AI's full reasoning lives off-chain; the on-chain memo is the relay of it.
A transaction is never paused once it is already on-chain — by definition, if it made it into a block, it already cleared. The check happens at one place only: at the sequencer, before the block is sealed.
Adaptive consensus threshold
The bar required to clear a transaction rises with its risk. A routine transaction clears against a light co-signing check; a higher-stakes one escalates to a full Byzantine-Fault-Tolerant (BFT) vote of the Consensus Council — a majority of an odd-numbered panel must agree (for example, five of seven). The threshold adapts to the transaction; the consensus process is the same.
What is metered
Each evaluation consumes GasIQ — the unit that meters AI evaluation the way gas meters computation — paid in the native $AIERC. It is a real consumed service: a real evaluation runs for a real charge.
On other chains: integration over RPC
AIERC protection reaches other EVM-compatible chains (EVM = Ethereum Virtual Machine, the common engine most chains run) through the AIERC RPC connection. A user points their wallet's network connection at the AIERC endpoint, and every outgoing transaction is evaluated before it is broadcast.
Two outcomes, and one hard stop
Over RPC there is no pause and no memo — those belong to Chaingentic alone. The connection layer informs and the extension stops the worst; neither can write to a chain it does not run. This is real protection and an open on-ramp — but the on-chain AI stays exclusive to Chaingentic.
What stays private
The published surface is the behavior: the verdicts, when they happen, and what the user sees. The off-chain decision-making is not published — how the AI weighs a transaction, the signals it reads, and the exact composition of the Consensus Council. The on-chain memo proves that the AI acted and anchors its conclusion; it does not expose how it decided.
Patent and licensing
On-chain AI — an AI evaluation gating a chain's own block production, with the AI's activity recorded on-chain — is the protected invention. Patent Pending: CIPO 3304029; WIPO PCT/IB2026/052205.
- Use it over RPC — the off-chain integration on other chains is the open on-ramp.
- Use it on Chaingentic — the full on-chain experience, native to our chain.
- Build on-chain AI into your own chain — that is the patented mechanism, and it requires a license.
The patent covers the off-chain AI reasoning, evaluation, orchestration, and verdict generation, and their use to gate on-chain settlement — not the tradable assets, the token, or ordinary market activity.