The cryptocurrency industry just took two parallel leaps toward a future where AI models, not humans, are the primary market participants. Within days of each other, Gemini and The Open Network (TON) unveiled infrastructure that lets artificial intelligence systems autonomously trade on centralized exchanges and manage on-chain wallets through Telegram bots.
Gemini’s Agentic Trading: On May 15, 2026, Gemini announced that its full trading API is now integrated with the Model Context Protocol (MCP), an open standard originally developed by Anthropic. Compatible AI agents—including ChatGPT and Claude—can now pull market data, query order books, place orders, and manage positions directly from user‑linked Gemini accounts. Users set budgets, strategies, and risk caps, while modular “Trading Skills” give agents access to dollar‑cost averaging, grid trading, multi‑leg structures, and volatility plays. The exchange stressed that human oversight remains through configurable limits, but the underlying shift is clear: a growing slice of resting and market orders on Gemini will originate from black‑box models, not human decision cycles.
TON’s Agentic Wallets: On April 28, 2026, TON Tech introduced Agentic Wallets, a standard that turns Telegram AI bots into semi‑autonomous financial entities. Each agent creates its own on‑chain wallet funded by the user but controlled under a split‑control architecture—the user holds master keys, while the agent gets narrow permissions for transfers, swaps, and DeFi interactions. No intermediary custodies funds. The system plugs directly into Telegram’s massive bot ecosystem, enabling users to ask a bot in chat to pay for services, trade tokens, or execute complex strategies without routing through custodians.
The contrast between the two approaches is stark. Gemini centralizes agentic activity inside a regulated, custodial CEX, treating AI as a new client type that humans merely configure. TON, on the other hand, pushes autonomy to the network edge via non‑custodial wallets on its Layer‑1 blockchain, with Telegram as the interface layer. Both, however, point to the same outcome: capital is becoming a semi‑autonomous process, and the next generation of market players will be swarms of un‑audited models systematically optimized around fee, tax, and compliance constraints.
Security concerns are escalating alongside the excitement. Gemini’s order flow now faces the risk of cascading feedback loops if many agents run similar strategies on overlapping data. TON’s design surfaces attack vectors like prompt‑injection, Telegram account takeovers, and poorly written agent logic that could autocompound bad positions. Legal liability for money laundering or DeFi exploits carried out by these agents remains undefined, making both launches a debate‑ready inflection point for regulators.
For crypto markets, the dual announcements signal a paradigm shift. AI is moving from a signal tool to a direct actor on order books and blockchains. The “HFT villain” of tomorrow may not be a quant firm but a swarm of AI models, and platforms are now building the rails for that reality.
