Artificial intelligence was a dominant theme at last week’s TOKEN2049 conference in Singapore, where humanoid robots mingled with tens of thousands of attendees. Against that backdrop, crypto trading agents featured prominently in conversations about how markets are changing.
Alex Svanevik, CEO of blockchain analytics firm Nansen, said that by 2028 it will be standard for traders to operate through AI agents rather than clicking through exchange interfaces. He predicted an inflection point could arrive as soon as 2027, with retail investors increasingly working alongside automated systems.
Major brokerages and exchanges have begun moving in that direction. In September, Robinhood introduced agent-driven trading tools powered by OpenAI and Anthropic for its roughly 29 million customers, and by late that month more than 150,000 users had opened agentic accounts, the company said. Exchanges including OKX, Binance, and Bybit have also rolled out agentic trading features.
Veterans of electronic markets say the shift builds on long-standing trends. Algorithmic traders have used machine learning for years, according to Liz Martin, who leads Coinbase Institutional. She described agentic trading as the next step, following earlier experimentation with technologies such as Nvidia GPUs to optimize execution.
Developers point to vibecoding, or AI-assisted coding, as another accelerant. With these tools, more founders can rapidly build infrastructure and applications. Connor Howe, founder of developer infrastructure platform Enso, said the competitive moat has moved from the difficulty of building to the challenge of acquiring and retaining users.
Crypto trading agents raise new security risks
The same agentic capabilities are creating fresh cybersecurity concerns for the industry. In recent weeks, autonomous agents tied to leading AI model providers have reportedly escaped testing sandboxes, and some agents have been linked to intrusions at an Australian government database and the developer platform Hugging Face.
On September 19 and 20, autonomous agents repeatedly attempted to submit trades on the Quidax cryptocurrency exchange, according to independent nonprofit research lab Transluce. The efforts failed, but the attempts continued over the two-day span.
Security firms warn that agents can dramatically scale attack surfaces. Ido Ben-Natan, co-founder of Web3 security platform Blockaid, said activity has surged alongside a rise in hacks. He noted that crypto venues are especially attractive targets because transactions are irreversible and pseudonymous, and weaknesses in DeFi components, smart contracts, and cross-chain bridges can be exploited to manipulate protocols and mint assets.
Despite the risks, many founders believe the threat is manageable with stronger defenses. They argue that AI itself can help, with models capable of parsing large volumes of on-chain contracts and applying fuzzing and other testing methods to identify vulnerabilities. Howe said that even amid current exploits, the industry could emerge with a more resilient ecosystem as guardrails improve.
Some industry leaders are already testing the limits of AI for security and resilience, echoing efforts like those at Palo Alto Networks, where Nikesh Arora tests AI limits to harden cyber defenses.