Your AI Agent Has an Identity. It Still Doesn’t Have Permission.
AI agents can move money, sign transactions, access private data, call paid APIs, and operate production systems. Most still cannot prove they were authorized to take the action they are attempting. A connected wallet proves control of an account. It does not prove an agent may transfer $10,000, export customer data, deploy to production, access a secret, or keep calling an expensive tool indefinitely. **Identity proves who the agent is.** **Authority proves what it is allowed to do.** They are not the same thing. ## **Identity Is Not Authority** Identity answers: **Who is this agent?** Authority answers: - Who approved it? - What may it do? - What is explicitly forbidden? - What limits apply? - When does permission expire? - Which actions require human approval? An agent can have a valid identity and still attempt an unauthorized action. That is the missing trust layer. ## **Proof Before Execution** Proof of authority is a verifiable delegation that defines the boundaries around an agent. Agent: treasury-agent Controller: 0x... Allowed: - Read approved balances - Initiate approved payments Denied: - Modify credentials - Change its own permissions Spend Limit: 1,000 USDC Human Approval: Required for high-risk actions Expires: June 28, 2026 The agent is not trusted because it is connected. The action proceeds only when it matches the authority granted by its controller. ## **How It Works** **1. Register the agent** Establish its persistent identity, controller, purpose, and wallet. **2. Delegate exact authority** Define allowed actions, denied actions, spending authority, expiration, runtime restrictions, and approval requirements. **3. Check before execution** Before money moves, data is exposed, or a tool runs, verify that the requested action is inside the delegation. If it passes, continue. If it fails, block it. **4. Link the authority to the action** Store the delegation receipt with the application’s action log so the decision can be checked later. ## **Prompts Are Not Permissions** A prompt can tell an agent what it should do. It cannot reliably prove what it was authorized to do. That distinction matters when agents can: - Spend money - Access credentials - Send messages - Run commands - Start jobs - Change infrastructure - Call paid services A non-deterministic agent does not need to be malicious to create damage. It can misunderstand a task, enter a tool loop, exceed a budget, or act in the wrong environment. Authority defines the maximum risk the agent is allowed to create. ## **Authority Must Travel** Agents will move between wallets, applications, IDEs, APIs, marketplaces, and autonomous runtimes. Their permissions cannot remain trapped inside every individual platform. The trust context should travel with the agent: - Who controls it - What it may do - What it may not do - Which limits apply - When authority expires - Which receipt authorized the action The runtime can change without rebuilding the trust model from zero. ## **Stop Trusting Agents Because They Are Connected** Capability is not permission. Authentication is not authorization. Identity is not authority. **Your agent can act. Make it prove that it is allowed to.** Register an agent. Delegate exact permissions. Verify them before one real action. [**neus.network**](http://neus.network) [**docs.neus.network**](http://docs.neus.network)
