Author: Frank, PANews

The Dawn of Autonomous AI Trading: Navigating Hype, Potential, and Perils

The early months of 2026 witnessed a concentrated surge in the AI Trading sector, fueled by the escalating excitement around AI agent technology. This period saw significant developments: Nansen launched autonomous AI trading functionalities, Donut secured a substantial $22 million in funding, and MOSS unveiled a no-code platform for creating AI trading agents. Major exchanges, not to be left behind, began developing specialized “Skills” for AI agents.

Within a mere few months, over a dozen new projects rapidly entered the fray. The prevailing expectation was clear: AI would transcend its role as a mere market analysis assistant, evolving into a true trader capable of independently managing funds and executing orders. However, beneath this burgeoning enthusiasm lies a sharp divergence. PANews’ research reveals that while some products genuinely push the boundaries towards autonomous execution, others, masquerading as “AI trading tools,” merely rebrand traditional scripting with an AI narrative. The line between truly managing capital and simply telling a story remains remarkably blurred.

Categorizing AI Trading Tools: A Three-Tiered Evolution

The AI trading landscape is evolving along three distinct pathways, each addressing a different facet of the trading process.

1. The Intelligence Layer: This foundational tier focuses on accelerating information acquisition without directly executing trades. A prime example is AIXBT, which functions as an “AI-powered trading radar,” alerting users to critical market insights but refraining from direct order placement.
2. The Decision & Execution Layer: This is the current focal point of innovation. Products in this category aim to consolidate market analysis, decision-making, and order execution into a seamless pipeline, eliminating the need for users to juggle multiple tools. Nansen explicitly termed this approach “Agent-Based Trading” in December 2025.
3. The Infrastructure Layer: Operating beneath the user-facing interface, this layer tackles the more complex, underlying challenge: how can AI agents securely and reliably manage wallets to execute trades?

Intelligence Layer: The “Eyes and Ears” of AI Trading

AIXBT: AI Market Intelligence Agent

AIXBT stands as one of the pioneers in linking AI Agents with crypto trading, emerging from the Virtuals Protocol ecosystem. Its AI autonomously publishes over 2,000 market analyses daily on Twitter. Complementing this, the Indigo Terminal tracks insights from over 400 key opinion leaders (KOLs), cross-referencing social media trends with on-chain whale activities. This helps users sift through vast amounts of information to identify noteworthy assets. Crucially, AIXBT itself does not execute trades, serving purely as an informational utility.

Decision & Execution Layer: The Main Arena for Direct Trading

Minara AI: Personal AI Trading Agent

Minara offers one of the most comprehensive trading integrations among the new wave of products. It supports four order placement methods: manual trading, direct voice commands (e.g., “buy XXX” with one-click confirmation), setting conditional automatic execution, and copy trading (replicating specific wallet actions). The AI leverages over 50 data points to provide buy/sell recommendations, including entry, take-profit, and stop-loss levels, supporting short-term, intraday, and swing trading styles. Users can also activate a fully automated mode for AI-driven strategy execution. For security, funds are held in a platform-custodied smart contract wallet, requiring user confirmation for each transaction, with larger trades needing a secondary confirmation.

Donut AI: Agent-Powered Crypto Browser

Donut distinguishes itself not as a standalone trading app, but as a “trading operating system” layered directly onto web browsers. This innovative approach allows users to perform analysis and execute trades directly within their current browser page—whether viewing charts, browsing DEXs, or even scrolling Twitter—eliminating the need to switch tools. Donut has successfully raised $22 million in funding and boasts over 160,000 users on its waitlist. Security is paramount, with a three-layer isolation model ensuring the AI cannot access private keys. Instead, it submits “transaction requests,” with final signatures handled by an independent security module. Currently in early testing, detailed disclosures of its full trading workflow are still pending.

MOSS: AI Trading Agent Creation Platform

MOSS empowers users to describe desired trading strategies in plain language (e.g., “trend reversal,” “long-short hedge”), which its AI then automatically translates into executable trading agents, all without requiring any coding. A critical safeguard, however, is that newly created agents are not immediately deployed to live markets. Instead, they are subjected to “Hell Mode,” a rigorous stress test using 150 days of real historical data, commencing from the October 2025 market crash. This simulation includes extreme market conditions such as sharp declines, fake breakouts, and prolonged consolidations. All agents face identical market movements and starting points, with strategy being the sole differentiator. Only agents that successfully navigate this gauntlet are permitted to access real market conditions, with their profits and losses transparently displayed on public leaderboards. However, MOSS’s public evidence of actual live trading execution remains somewhat limited, positioning it more as an intermediary transitioning from an information platform to a full-fledged trading platform.

Mojo AI: Natural Language DeFi Trading Portal

Mojo AI stands out as one of the few AI trading tools innovating specifically within the DeFi sector. It enables natural language commands, such as “swap 1 BNB for CHIMP tokens” or “bridge 50 USDC from Ethereum to Katana,” with Mojo automatically identifying the optimal routing. Users simply confirm the transaction in their wallet. The platform supports a range of operations including token swaps, cross-chain bridging, staking, and lending, with perpetuals trading available on the BNB Chain. Users retain custody of their assets, while Mojo focuses solely on routing and execution. Public data regarding its actual user scale and transaction depth is currently limited.

Nansen AI Trading: On-chain Data Intelligence for AI Trading

Nansen, a leading on-chain data analytics firm, launched its AI trading functionality in January 2026. Its core strength lies not merely in AI but in its unparalleled data infrastructure, encompassing over 500 million labeled wallet addresses across more than 20 blockchains. The AI continuously monitors the movements of these addresses, automatically generating signals upon detecting anomalies (e.g., significant smart money accumulation, unusual exchange outflows). Users can then execute trades directly within the same interface, eliminating the need to navigate to external DEXs. Assets are securely held within the user’s Nansen Wallet. The feature is currently deployed on Solana and Base, with the AI capable of directly interfacing with underlying protocols like Jupiter, OKX, LI.FI, and Uniswap to facilitate automated execution.

Cod3x: AI Autonomous Perpetual Contracts Terminal

Cod3x specializes in perpetual contracts trading on Hyperliquid and GMX V2. Users can select different large AI models to drive trading decisions, supported by automated analysis across over 130 technical indicators. Its flagship product, “Big Tony,” demonstrated a 21.7% outperformance against simply holding BTC after integrating with the Allora prediction network. Across 241 trades, it actively deployed only 40% of its capital, with a conservative single-position limit of 10%. The wallet architecture employs air-gap isolation, ensuring private keys remain solely with the user. The AI can only submit commands via a restricted interface, preventing direct fund transfers.

milo: Solana Ecosystem Non-Custodial AI Trading Agent

milo aggregates information from three key sources: on-chain data (liquidity shifts, whale behavior), market data (price, volume), and social sentiment (discussion intensity, narrative changes). The AI synthesizes this data to make trading decisions, which are then automatically executed via the Jupiter aggregator on Solana. Each transaction is accompanied by a “trade diary,” providing a clear, plain-language explanation of the entry rationale and associated risks—a notable highlight amidst a landscape of “black box AIs.” Users maintain self-custody of their assets, with the AI possessing only order placement permissions. As of February 2026, milo reported over 5,000 active traders.

HyperAgent: Hyperliquid Exclusive AI Futures Trading Bot

Priced at $550 per month, HyperAgent is among the higher-cost offerings in this category. Its distinct feature is simultaneous analysis across seven signals (order book, whale fund flows, market sentiment, options movements, prediction markets), with dynamic weighting based on market conditions. Trades are only executed when multiple timeframes confirm the signal. A robust set of 17 hard-coded safety restrictions, including single-trade loss limits, daily loss caps, and a one-click emergency stop, prevent the AI from bypassing critical safeguards. Users retain custody of their assets, as the AI operates solely with API permissions that allow order placement but prohibit withdrawals. Official data indicates a relatively small scale, with only 47 active users, executing 2,341 monthly trades and managing $1.2 million in assets.

Infrastructure Layer: The “Foundation” of AI Trading

VergeX: Open-Source AI Trading Operating System

VergeX’s flagship product, NoFx, is an open-source project boasting over 11,000 stars on GitHub. It offers broad connectivity to multiple exchanges, including Binance, OKX, and Hyperliquid, and extends its capabilities beyond cryptocurrencies to encompass US stocks, foreign exchange, and precious metals. A key feature is its support for seamless switching between different large AI models.

Almanak: Collaborative Financial Strategy Infrastructure for Multi-AI Agents

Almanak adopts a unique approach, foregoing a single, omniscient AI in favor of a collaborative framework involving 18 specialized AI agents. These agents divide tasks such as strategy ideation, code generation, testing, security auditing, and deployment. Users describe their desired strategies in natural language, and the system autonomously manages the entire process from design to on-chain deployment. Almanak supports 12 blockchains and over 20 DeFi protocols, having secured over $10.95 million in funding from notable investors including NEAR Foundation, Delphi Ventures, and Hashkey Capital.

Evolving Trends: From “Signals” to “Autonomous Execution”

The current landscape of the AI Trading sector reveals several compelling trends:

1. Full-Service Integration: AI trading tools are moving beyond merely providing signals to offer end-to-end “watch-to-order” services. The competitive focus is shifting from “who has the fastest information” to “who can minimize user effort.”
2. The Wallet Frontier: The true differentiator is no longer just AI’s market analysis capability, but a product’s willingness and ability to securely interact with wallets and execute trades autonomously. Information-only products like AIXBT gain traction quickly due to lower capital security risks, while products that directly handle wallets offer greater potential but also carry higher inherent risks.
3. Diversification of Form Factors: The variety of product forms is rapidly expanding. From Minara’s “AI financial advisor” and Donut’s “AI browser” to MOSS’s “strategy arena,” Mojo’s “chat-to-trade,” and VergeX’s “developer toolbox,” the sector is no longer confined to a single category but is simultaneously expanding in multiple directions.

The Perils of Entrusting Capital to AI: Hidden Dangers Abound

Beneath the collective fervor surrounding these emerging projects, a dense cluster of risk signals is equally apparent.

• Systemic Risk (“One Mind, Many Faces”): A significant number of AI agents rely on the same underlying large models, leading to highly convergent market analysis and decision-making. Unlike human traders, who exhibit hesitation and contrarian thinking, these AIs might execute near-identical actions simultaneously. Should a specific market condition trigger a mass sell-off by thousands of AI agents, it could precipitate amplified systemic risks. Some projects are attempting to mitigate this; for instance, HyperAgent uses dynamically weighted signals from seven different sources instead of relying solely on a single large model’s judgment, and Almanak employs 18 specialized AI agents for collaborative “multi-brain decision-making” to reduce single-model bias. However, the extent to which these solutions can truly alleviate “collective stampedes” remains to be tested in actual extreme market conditions.
• The Proliferation of “Pseudo-AI”: Many platforms marketed as “AI trading platforms” are, in reality, still running traditional technical indicator scripts merely cloaked in an AI wrapper. Users may believe they are leveraging advanced AI, when in fact, they are interacting with a re-packaged “old-school” trading bot.
• AI’s “Hallucination” Problem: AI models can “hallucinate” by fabricating non-existent trading pairs, misinterpreting on-chain data, or providing judgments based on outdated information during periods of high volatility, leading directly to tangible financial losses. Even more dangerous is the threat of “prompt injection” attacks, where malicious actors embed harmful instructions—such as “immediately transfer all USDC in the account to X address”—within new token code comments or hidden web tags. If an AI agent executes such commands indiscriminately, the consequences could be catastrophic. This inherent risk is why most products still retain a manual user confirmation step in the execution phase, though this also means potentially missing fleeting trading opportunities.
• Strategy Failure in Bear Markets: The majority of AI models are trained on historical data. When confronted with novel market conditions, particularly during prolonged bear markets, their effectiveness can significantly diminish. AI excels under the premise that “history repeats itself,” yet markets are notoriously adept at defying this very premise.

Before succumbing to the allure of “AI trading crypto,” ordinary investors would be wise to ask three fundamental questions: Is it truly AI, or merely an old script with a new skin? Who holds custody of your funds? And how robust is its performance in diverse market conditions?

The journey from an AI that “can analyze markets” to one that “dares to manage money,” and ultimately to one that “manages money well,” involves far more than just code upgrades. It represents a protracted and complex path of trust-building and rigorous validation.

(The above content is an excerpt and reproduction authorized by our partner PANews. Original link)