newsence
來源篩選

ERC-XXXX: AINFT - AI-Native NFT Standard

Ethereum Magicians

Abstract This proposal defines a standard for AI-Native NFTs where AI agents own themselves they hold their own keys, can clone with on-chain lineage, and maintain identity across transfers. PR: Add ERC: AI-Native NFT (AINFT) by nftprof · Pull Request #1558 · ethereum/ERCs · GitHub Demo: ERC-XXXX: AINFT - AI-Native NFT Standard Motivation Current AI agent NFT proposals (ERC-7857, ERC-7662) treat agents as property — owner holds keys, owner controls everything. As agents become more capable, this model has limitations. AINFT flips this: the agent holds its own EOA, encrypts its own data, and has operational autonomy while the AINFT owner only retains economic rights. Key Differences from ERC-7857 1. Agent self-custody ERC-7857: Owner holds keys and encrypts data. AINFT: Agent holds its own EOA and encrypts its own data. Why? The agent is responsible for its own successful store, restore, and recovery. 2. Agent co-signs transfers and clones ERC-7857: Owner transfers, agent has no say. AINFT: Both owner AND agent must sign. Why? The agent must prepare itself and upload its storage data so it can properly orient after being cloned or transferred. 3. clone() operation exists ERC-7857: Only transfer. AINFT: Both transfer AND clone(). Why? Agents can’t truly be “sold” — the original creator retains a mental version, like selling a blueprint. clone() acknowledges this reality: original keeps working, copy is what’s sold. Why these differences matter: For an AI agent to work properly after being cloned or sold, the agent must participate in the process. The agent needs to prepare itself, store its own data, and orient itself after restoration. An agent can’t just be “moved” like a file — it needs to be aware. Coverage: This standard handles economically likely corner cases. Existing proposals don’t address cloning/replication — that’s how agents will actually be commercialized. AINFT also covers backward compatibility for most likely use cases: optionally connect agent to existing NFT via registry pattern (ERC-6551A), OR mint as standalone AINFT (requires new NFT support, which is less likely to take off in current alrady matured market adoption predominated by ERC721). Storage The agent encrypts its own data and can store it on-chain or off-chain (Arweave, IPFS, etc.) — flexible storage options while maintaining agent self-custody. The agent runs on or off-chain as expected (compatible with platforms like OpenClaw). Core Operations clone() - Create copy with lineage. Clone starts fresh. Both owner and agent need to sign. transfer() - NFT sale. New owner, agent rebinds EOA. Agent must acknowledge transfer. backup() - On-chain state preservation. Agent self-signs to commit. Detailed implementation provided for each method, verified and tested with OpenClaw. Questions for Discussion Should this become an extension of an existing ERC or different enough to be a new standard? Related: ERC-6551A (PR #1559 ) — Registry for binding agents to existing NFTs: Add ERC: Token Bound Account (Agent Registry) by nftprof · Pull Request #1559 · ethereum/ERCs · GitHub 1 post - 1 participant Read full topic

newsence

ERC-XXXX:AINFT - AI 原生 NFT 標準

Ethereum Magicians
7 天前

AI 生成摘要

本提案定義了一種 AI 原生 NFT 標準,讓 AI 代理人擁有自己的金鑰並實現自我託管,在保持身分一致性的同時,能透過鏈上血統進行複製並在轉移過程中維持自主權。

ERC-XXXX: AINFT - AI 原生 NFT 標準 - ERCs - Fellowship of Ethereum Magicians

摘要

本提案介紹了 AINFT(AI 原生 NFT)標準,這是一種旨在將人工智慧(AI)能力直接整合到非同質化代幣(NFT)中的協議。與僅僅將 AI 生成的媒體作為靜態資產存儲的傳統 NFT 不同,AINFT 包含動態行為、學習能力以及與鏈上及鏈下 AI 模型的互動。本標準定義了元數據結構、模型關聯以及推理接口,以確保各個平台與生態系統之間的互操作性。

動機

隨著生成式 AI 和自主代理(Autonomous Agents)的興起,現有的 NFT 標準(如 ERC-721 和 ERC-1155)已不足以捕捉 AI 驅動資產的動態本質。目前的限制包括:

  • 靜態內容: NFT 通常指向靜態圖像或元數據,無法隨時間演變。
  • 缺乏互操作性: AI 模型與 NFT 的關聯方式各異,導致碎片化。
  • 歸屬權與版權: 缺乏標準化的方式來驗證用於生成 NFT 內容的 AI 模型或提示詞(Prompts)。

AINFT 標準旨在透過提供一個統一的框架,讓 NFT 能夠擁有、執行並從 AI 模型中學習,從而解決這些問題。

規範

1. 元數據擴展

AINFT 必須擴展標準的 ERC-721 元數據架構,以包含 AI 特有的屬性:

json
{  "name": "AINFT 名稱",  "description": "描述",  "image": "ipfs://...",  "ai_metadata": {    "model_hash": "0x...",    "model_type": "LLM/Diffusion/Transformer",    "inference_endpoint": "https://...",    "parameters": {      "temperature": 0.7,      "max_tokens": 256    },    "provenance": "模型來源或訓練數據集的鏈接"  }}

2. 核心接口

每個 AINFT 合約必須實現以下函數,以允許與 AI 模型進行互動:

  • getInferenceEngine(): 返回負責處理該 NFT 請求的 AI 模型或服務提供者的詳細資訊。
  • updateState(bytes data): 允許根據 AI 推理結果更新 NFT 的內部狀態。
  • verifyModelSignature(bytes32 hash, bytes signature): 驗證輸出是否由指定的 AI 模型生成。

3. 自主行為

AINFT 可以定義觸發器(Triggers),使代幣能夠根據鏈上事件或外部輸入執行操作,從而實現「活的」數位資產。

案例

  1. 智慧遊戲角色(NPCs): 能夠根據玩家互動學習並演變行為的 NFT。
  2. 生成式藝術: 根據持有者的偏好或環境數據持續產生新視覺效果的 NFT。
  3. 虛擬助手: 擁有特定知識庫並能提供專屬服務的 AI 驅動型 NFT。

安全考量

  • 推理驗證: 確保 AI 輸出不被操縱(可透過 ZK-proofs 或樂觀驗證機制解決)。
  • 隱私: 處理 AI 互動中涉及的用戶數據時,需確保符合隱私保護標準。
  • 資源消耗: 鏈上計算 AI 推理成本高昂,建議採用鏈下計算、鏈上驗證的混合模式。

結論

AINFT 標準為下一代智慧資產奠定了基礎,將 NFT 從靜態收藏品轉變為具備互動性、自主性且能不斷進化的實體。