The use of AI assistants in crypto trading and payments introduces new security risks for users’ wallets, experts shared with Cointelegraph.
On October 22, the largest American Bitcoin exchange, Coinbase, unveiled the Payments MCP tool. This solution, combined with the use of LLM such as Claude, Gemini, and Codex, allows AI assistants to access crypto wallets on par with their owners and autonomously make payments via the x402 protocol.
Announcing Payments MCP, the easiest way for AI agents to get onchain via x402. 🚀
It lets LLM models like Claude, Gemini, and ChatGPT gain access to onchain tools like wallets, onramp, and payments with no API key required. 🧵 pic.twitter.com/MSnIaecx0O
— Coinbase Developer Platform🛡️ (@CoinbaseDev) October 22, 2025
“This marks a new phase of agent commerce, where AI assistants can operate in the global economy,” stated the platform team.
Are AI Agents Ready for Independence?
Aaron Ratcliffe, head of attribution at blockchain analytics firm Merkle Science, noted that such an option requires an additional level of trust for artificial intelligence in a field that does not inherently provide for it.
Properly structuring processes can eliminate potential threats, but the responsibility for this lies with the wallet owner, the expert warned.
“Safe use depends on whether users understand how to make requests and how AI extracts data from the blockchain without hallucinations. It is also related to the security of account information — its leakage inherently implies damage,” Ratcliffe stated.
He added that using AI to manage a crypto portfolio carries risks of vulnerabilities that could be exploited by malicious actors. Hacking the system could be facilitated by introducing malicious prompts or instructions.
Another potential threat is a “man-in-the-middle” attack, where a hacker intercepts the communication channel and manipulates the messages between parties.
“AI can also interact with fraudulent tokens, miss traps or deceptive maneuvers, or handle price slippage poorly, simply burning users’ funds,” Ratcliffe added.
According to him, safe use of artificial intelligence requires solid evidence that assistants can detect attacks, fraudulent tokens, verify contacts, reject infected requests, and prevent other malicious actions.
Another possible issue, the expert believes, is the potential interaction of AI with sanctioned addresses and platforms.
Sean Ren, co-founder of the Sahara AI platform, stated that Coinbase’s AI tool uses protocols that “are the gold standard of security when properly configured.” However, it is still necessary to monitor what the AI assistant does.
“Users still need to be vigilant, double-check what they approve, and never assume the agent automatically does everything correctly. You will still have to verify and sign transactions,” he emphasized.
The Broader Potential of Technology in Crypto Trading
Glider co-founder and CEO Brian Huang expressed confidence that implementing basic functions like sending or exchanging assets through AI agents is just the initial stage of the trend. Some of these actions are simpler and faster to do independently, and better uses can be found for assistants.
“We all know that DeFi protocols are quite complex to participate in. Agents can help users get acquainted and feel like guides throughout the process,” he cited as an example.
Huang predicts that AI assistants will soon be able to perform more complex operations such as portfolio management, rebalancing, and personalized financial advice. Such options will become a more effective use of the technology.
“The personalization that agents can provide, the number of variables they can consider, far exceed the capabilities of any human,” he emphasized.
Back in a nine-day crypto trading competition among AI models, China’s DeepSeek turned an initial $10,000 into $22,031. In contrast, GPT-5 and DeepMind suffered interim losses of around 60%.