TL;DR
- The Open Platform launches "agentic wallets" on the TON blockchain — wallets controlled by AI without manual approval per transaction
- The standard is open source and builds on Telegram's network infrastructure
- Coinbase launched similar technology in February 2026; Stripe followed with its Machine Payments Protocol in March 2026
- Security experts warn: delegating financial control to AI agents opens new and serious attack surfaces
AI Takes Control of Crypto Wallets
The Open Platform (TOP) — the organization that built Telegram's official crypto wallet — has launched what they call "agentic wallets" on the TON blockchain. According to The Defiant, this is an open-source standard that enables AI agents to independently manage dedicated on-chain wallets, without the user having to sign or approve each individual transaction.
In other words: you define the frameworks, and the AI agent operates within them — on its own.
Agents are software with delegated authority. The core risk is the misuse of that authority.
What Exactly is an "Agentic Wallet"?
Traditional crypto wallets function as passive storage places where the user approves all actions themselves. An agentic wallet turns this on its head: an AI agent is assigned a dedicated wallet with predefined rules and can then act autonomously — buy, sell, pay — within these limits.
Among other things, the technology enables automated portfolio management, direct payments between machines, and the execution of smart contracts without human intervention.
A Rapidly Moving Market
TOP's launch occurs in a competitive field that is already in full swing. Coinbase launched its own agentic wallets in February 2026, with programmable spending policies and permission-controlled execution. Stripe followed up on March 18, 2026, with its Machine Payments Protocol (MPP) — an enterprise-focused standard for machine-to-machine payments with OpenAI and Anthropic as design partners.
The total market for AI-driven crypto wallets is estimated at around $7 billion, according to industry analyses. However, the actual on-chain volume from autonomous agents is far more modest: after filtering out self-trading wallets, real agent volume is estimated at around $1.6 million.
Security Researchers Are Not Reassured
The technology is promising, but security experts point to significant risks associated with granting AI agents financial autonomy.
Marco De Rossi, AI Lead at MetaMask, precisely formulates the core problem: agents are software with delegated authority, and the risk arises when that authority is misused — either through external manipulation or by the agent acting in ways its owner never intended.
A compromised AI system can, according to research from 2026, poison 87 percent of decisions in connected systems in just four hours. NIST warns in its 2026 report that AI agent systems introduce risks that go far beyond traditional software vulnerabilities — including so-called "prompt injection" attacks, where malicious instructions are tricked into the agent's decision-making basis.
The Most Concrete Threats
Among the security concerns highlighted in research communities are:
- Privilege creep: Agents are often given more access than necessary and can escalate their own rights over time
- Prompt injection: Attackers can submit manipulated instructions that override the agent's original goals
- Memory poisoning: False information is injected into the agent's persistent memory, affecting future decisions
- Compromise of private keys: Where agents are given access to keys, they constitute a significant attack surface
What Happens Next?
TOP's open-source approach differs from competitors' more closed solutions and can lower the threshold for developers to build their own agentic wallet integrations on the TON network — which inherently has access to Telegram's massive user base.
How quickly actual adoption will come is uncertain. Research communities agree that the technology is promising, but that trust, traceability, and human oversight are crucial prerequisites for autonomous financial agents to function safely in practice. Over 70 percent of businesses using agentic AI report positive returns — but this primarily applies to controlled enterprise environments, not open financial networks.
In a market where crypto theft reached $3.4 billion in 2025, the question of security is anything but academic.
