[Bitcoin Now] In the US, AI Invests Autonomously... South Korea Still Has a Long Way to Go

Robinhood Launches Agentic Trading
Virtual Asset Transactions to Be Supported in the Future
South Korea Lags Behind with Unresolved Regulations

In the United States, technology has emerged that allows artificial intelligence (AI) to autonomously invest in virtual assets in the future. In contrast, South Korea is being evaluated as technologically lagging behind, due to the lack of updated financial regulations and unresolved legal issues.

Robinhood launched its Agentic Trading and Agentic Credit Card services on May 27, where an AI agent performs trading and credit card payments on behalf of customers. The trading feature is currently offered as a beta service and supports only stock trading. However, the company plans to expand support to options, virtual assets, and futures trading in the future. This feature allows users to connect AI agents such as Anthropic’s Claude to a dedicated investment account, enabling the agent to execute trades based on instructions. Users can instruct the agent to detect loss risks through portfolio analysis, monitor promising sectors, or invest like a private equity firm by analyzing enterprise value data. For credit card payments, the agent handles everything from price searches and inventory monitoring to the final payment during the product purchase process.

Both services are based on a structure where the agent interprets the customer’s intent and executes transactions that meet specific conditions, without requiring direct customer action. Coinbase, the largest virtual asset exchange in the United States, also developed a payment infrastructure (x402 protocol) last month, enabling an AI agent to perform small payments using stablecoins over the internet without creating a separate account or payment procedure. Hyunkyung Yang, a researcher at iM Securities, said, “Both services commonly signal a transition from a structure where humans directly click the payment button to one where agentic AI determines the conditions and executes the payment.”

According to Yang, in this era of agentic AI, securing the customer’s power of attorney becomes more important than just acquiring customers. She stated, “It could become more important for financial companies to be granted payment and trading rights by the customer’s AI, rather than simply focusing on which financial application the customer opens.”

Google Gemini Generated Image

Robinhood Also Provides Solutions to the 'Power of Attorney' Issue

However, a key issue may arise around holding humans responsible for transactions executed by AI acting as their agent. For example, it is unclear how responsibility will be assigned when an AI incurs losses or violates its duty of loyalty as an agent.

To address this, Robinhood has implemented various mechanisms such as disclaimers and asset segregation. The service terms explicitly state, “AI-based transactions involve significant risks, including the possibility of losing the entire investment,” and include a disclaimer that AI performance is not guaranteed. Additionally, the AI agent can only access the agentic account, which is established separately from the main portfolio, ensuring that main assets remain protected even if the AI makes mistakes. Each AI-executed trade is accompanied by push notifications and a real-time activity feed, and the company can immediately cut off AI access if any abnormal signs are detected.

South Korea May Fall Behind Due to Mounting Regulatory Barriers

In South Korea, the virtual asset exchange Upbit has introduced a similar service, but it is still far from the agentic AI services available in the United States. On May 22, Upbit launched ‘Upbit Skill,’ which enables AI agents to use external AI to check prices and account information, process order requests, and perform deposit and withdrawal-related queries. When a user enters a natural language request, the agent can propose command structures or assist with execution.

In the South Korean virtual asset market, ambiguous legal regulations create an environment where it is difficult for operators to develop related technology. To enable autonomous AI trading, a company must first acquire a discretionary investment management license under the Capital Markets Act, because the AI would be managing assets without the customer’s confirmation for each transaction. However, virtual assets are not included in the range of eligible assets for discretionary investment management. Even if an operator obtains the license, it cannot pass the KOSCOM Robo-Advisor Testbed, which is required to commercialize agentic AI services. The testbed requires operators to test with real data for several months to verify normal operation, but virtual assets are not part of the testbed’s eligible assets.

Virtual asset exchanges also face difficulties in development due to potential conflicts of interest. If the exchange not only brokers trades but also manages customer funds, there is a risk that the AI could be used to artificially increase trading volumes by forcing unnecessary virtual asset transactions. An industry insider said, “If AI guides or skills provided by exchanges go beyond simple order assistance and even slightly involve recommending specific assets or acting as a proxy manager, there is a high risk of violating regulations for unregistered virtual asset businesses or unlicensed discretionary investment management, creating major obstacles to service enhancement.”

This content was produced with the assistance of AI translation services.