Agentic AI is set to transform user interactions with crypto wallets in trading and payments. While blockchain and AI executives believe in its safety, new risks are anticipated.
Recently, Coinbase introduced its tool, Payments MCP, enabling AI agents to use the same onchain financial tools as humans.
Announcing Payments MCP, the easiest way for AI agents to get onchain via x402. 🚀
It allows LLM models like Claude, Gemini, and ChatGPT to access onchain tools like wallets, onramp, and payments without needing an API key. 🧵 pic.twitter.com/MSnIaecx0O
— Coinbase Developer Platform🛡️ (@CoinbaseDev) October 22, 2025
With an LLM such as Claude, Gemini, or Codex, the tool allows them to access crypto wallets and execute payments autonomously, as stated by the Coinbase Developer Platform in a statement.
AI agents equipped with Payments MCP can pay for services, compute and retrieve paywalled data, tip creators, and handle certain business operations via the x402 protocol, an open, web-native payment system enabling instant stablecoin transactions, according to the Coinbase Developer Platform.
“This signifies a new phase of agentic commerce where AI agents can participate in the global economy,” noted the Coinbase Developer Platform.
Agentic AI in crypto can be secure
Aaron Ratcliff, the attributions lead at Merkle Science, commented to Cointelegraph that allowing an AI agent access to a wallet adds a trust element to a system designed to be trustless.
If well-constructed, it can be secure, but Ratcliff emphasizes that users ultimately hold the responsibility for safety.
“Safety hinges on users who know how to prompt and on the AI accurately pulling blockchain data without hallucinations. It also relies on the secure management of trading credentials; if those credentials are compromised, the damage is done.”
AI in portfolios may incur additional security risks
An April survey by CoinGecko of 2,632 crypto users found that many are open to AI trading on their behalf; 87% expressed willingness to let AI agents manage a portion of their crypto investments.
Ratcliff warned of potential security risks that malicious actors could exploit when employing AI within a portfolio. Instruction injection could be a means to hijack the system.
A man-in-the-middle attack, where an intruder stealthily intercepts communications to steal data, could also misdirect trades.
“The AI might similarly engage with scam tokens, miss honeypots or rug-pulls, or manage slippage poorly, risking users’ funds,” Ratcliff stated.
“I’d require evidence that the AI can handle front-running, apply slippage limits, identify scam tokens, and audit contracts in real-time before executing trades. It should also sandbox prompts, prevent injection, and block unauthorized access.”
Moreover, Ratcliff believes that compliance gaps might result in issues, like the risk of an AI sending funds to a sanctioned entity or exchange.
Even with safeguards, vigilance is necessary
In a conversation with Cointelegraph, Sean Ren, co-founder of Sahara AI, remarked that Coinbase’s tool employs model context protocols, “which are the gold standard for security when implemented correctly.”
“They function as a gatekeeper between the AI model and your wallet. The agent can only perform specific, approved actions—like checking balances or preparing payments for your confirmation—rather than autonomously reallocating funds or altering wallet settings,” he explained.
“Such actions are intentionally restricted, so even if someone attempts to deceive the AI via prompt injection, it cannot execute a transaction independently,” Ren continued.
However, Ren cautioned that safer does not mean infallible, and users must remain aware of the actions that the AI agent undertakes in their portfolio.
“Users must stay vigilant, verify what they approve, and never assume that the agent is automatically making the right decisions. Reviewing and signing transactions remains essential.”
AI agents are in the early stages
Brian Huang, co-founder and CEO of Glider, a platform for AI-driven crypto portfolio management, told Cointelegraph that basic functionalities, such as sending, swapping, and lending, are a good starting point for agents, yet the field remains nascent.
Related: AI agents require crypto to function in financial markets: Coinbase exec
“These straightforward actions can be completed with a click—you’re not asking ChatGPT to Venmo your friends, right? Many actions with agents demand more time,” he indicated.
“Agents function more like assistants; everyone knows that DeFi is too intricate for casual participation. These agents can aid users in onboarding and navigating the process.”
Huang anticipates that more advanced functionalities like portfolio management, rebalancing, and tailored financial guidance will emerge and prove to be more effective applications.
“The customizability that agents offer, with the multitude of factors they can assess, greatly surpasses what any human can offer,” he emphasized.
Magazine: How do the world’s major religions perceive Bitcoin and cryptocurrency?
