TON’s new Agentic Wallets standard lets Telegram AI bots hold user‑funded wallets on TON, spending within tight limits as semi‑autonomous financial actors inside chat.
Summary
- TON Tech has launched “Agentic Wallets,” an open, self‑custodial standard that lets AI agents on Telegram hold funds and execute on‑chain transactions on the TON blockchain without per‑action user approval.
- Each agent gets a dedicated wallet funded and owned by the user, with hard spending limits and revocable access, effectively turning bots into bounded financial actors that can trade, pay subscriptions, and interact with DeFi inside Telegram’s roughly 1 billion‑user ecosystem.
- The move is being pitched by TON Tech’s Andrew Grekov as the shift from “assistants to actors,” but it also opens a new attack and governance surface around agent misbehavior, prompt‑injection, and blurred liability between users, developers, Telegram, and the TON network.
TON Tech — the infrastructure team behind The Open Network — rolled out Agentic Wallets on April 28, 2026, describing them as “self‑custody wallets designed for autonomous AI agents on TON” that finally give Telegram bots a native way to move money. According to TON’s docs and supporting announcements, each AI agent can spin up its own on‑chain wallet, funded directly by the user; the agent then manages that balance autonomously, while ownership remains anchored to the user’s main wallet and can be revoked at any time.
TON’s AI agents get real wallets, not just UX gloss
Crucially, this is not a custodial layer or a centralized key‑escrow hack. TON Tech stresses that “no intermediary holds funds at any point” and that existing TON wallets require “no upgrades” to plug into the scheme, which is implemented as a standard contract pattern rather than a new app silo. The design uses a split‑control architecture: users keep the master keys; agents get narrow, contract‑level permissions to initiate transfers, swaps, and DeFi interactions within a predefined budget, with the ability for users to pull funds or kill the agent’s access at will.
From a product perspective, the key move is that Telegram itself becomes the UI and distribution layer. Telegram’s bot infrastructure and bot‑to‑bot messaging already run across a reported 1 billion‑plus users; Agentic Wallets plug into that fabric so that a user can literally ask a bot in chat to “create a wallet,” fund it, and then let it pay for services, exchange tokens, or execute transactions from inside the same interface. As Grekov puts it, “Agentic Wallets turn AI agents from assistants into actors — agents on Telegram can not only communicate, but transact,” collapsing the distance between conversation and settlement into a single app.
Programmable capital with a will — and a bigger attack surface
The concrete use cases TON Tech and third‑party write‑ups are pushing are exactly the ones you’d expect: trading bots with predefined budgets, DeFi agents that handle staking and portfolio rotations, and automation for subscription payments, API usage and micro‑transactions, all without routing through custodians. Blockster’s analysis is blunt: this “pushes Telegram‑based AI agents beyond simple assistants and into something closer to autonomous financial actors,” meaning that once budgets and rules are set, the agent can hold balances, make payments, and interact with on‑chain applications without a human clicking “confirm” on every transaction.
For crypto, that is the actual “AI + blockchain” crossover that matters: not vaporous “AI tokens,” but agent frameworks that can maintain positions, roll perps funding, dollar‑cost average into a basket, or run a Polymarket/Kalshi‑style prediction‑market book 24/7 inside a chat app. In practice, it means your next trading strategy, recurring remittance flow, or cross‑border bill‑pay could be delegated to scripts with persistent identity and direct on‑chain reach, turning capital into something closer to a semi‑autonomous process than a pile of passive balances.
The flip side is that the governance and security surface just exploded. None of the launch materials resolve what happens when an “agent” griefs a protocol, front‑runs retail flow, or becomes part of a cartel coordinating MEV‑style behavior across DeFi inside Telegram. The attack vectors are obvious: prompt‑injection or jailbreaks that subvert an agent’s policy layer, Telegram account takeovers that let an attacker reconfigure or drain agent wallets, or poorly written agent logic that autocompounds bad positions and nukes user balances while formally staying “within budget.”
Legally and politically, the liability chain is completely undefined: when an agent running in Telegram uses an Agentic Wallet to launder funds or exploit a DeFi contract, the blame can be projected onto the user, the bot developer, TON Tech’s standard, or Telegram’s distribution layer, with no clear doctrine yet to apportion responsibility. That ambiguity is exactly why this launch is bigger than another “AI wallet” gimmick — it’s the first serious attempt to normalize autonomous agents as on‑chain actors inside a mainstream consumer app, with all the upside and all the systemic risk that implies.
