Artificial intelligence has transformed from institutional back-office tool to retail investment technology, with 62% of US retail investors now using AI in their investment process according to a March 2026 survey of 938 participants. However, the adoption reveals a clear pattern where 65% of AI users report improved results, though this remains a self-reported figure rather than risk-adjusted performance data. As reported by Investing.com, investors demonstrate caution by trusting AI analysis only after verification elsewhere, with 53.5% of users trusting AI analysis only after verifying it elsewhere, and only a small minority handing over execution to automated systems. The dominant use case remains research activities such as news summarization and stock screening, while misleading recommendations rank as the top concern among investors. Practitioners reinforce this gap, with Kieran Garvey, AI research lead at the Cambridge Centre for Alternative Finance, noting that the technology for delivering financial advice remains "nowhere near reliable" today.
The AI investing landscape encompasses three distinct technologies with varying maturity levels. The first category includes large language models like ChatGPT, Gemini, or Claude, which predict plausible text with equal fluency regardless of accuracy and represent a possible source of hallucination. The second category consists of robo-advisors that automate narrow, measurable tasks such as rebalancing and tax-loss harvesting, which are mature and reliable precisely because their objectives are defined. The third and most controversial category involves agentic AI models granted permission to execute live trades through brokerage APIs in continuous observe-decide-execute loops. According to Investing.com, agentic AI represents the newest, least proven layer where both retail enthusiasm and regulatory concern are concentrated.
AI trading competitions have revealed significant performance gaps between theory and practice. Across six competitions covering equities, ETFs, crypto, and prediction markets, most produced at least one profitable model, but only two delivered profitable median results. A major contest provided eight frontier models with $10,000 each to trade US tech stocks for two weeks, resulting in a combined portfolio loss of about one-third of its value. As reported by Investing.com, only six of 32 runs turned profitable, with models making between 158 and 1,418 trades, highlighting the need for robust engineering systems for risk management. The results underscore that typical AI traders still lose money despite occasional standout performers, with performance turning only after adding risk controls, automated stop-losses, position limits, and forward testing.
Major platforms have adopted distinct strategies for AI integration, with most focusing on human-in-the-loop approaches. Public's assistant operates as a copilot allowing natural language analysis of companies and financial data, while Robinhood's Cortex provides portfolio analysis and market insights. eToro's Tori extends similar functionality for its Popular Investors program. The more aggressive approach involves agentic execution, where Robinhood allows users to connect external systems like ChatGPT through the Model Context Protocol for automated trading without per-transaction confirmation. As reported by Investing.com, platforms emphasize tighter guardrails and human oversight rather than raw AI capability, with Webull's president framing this as 'zero commission 2.0' for the trading industry. Robinhood confines agents to a segregated wallet's pre-loaded balance, requires preview approval on some trades, notifies on every transaction, screens for suspicious activity and allows instant disconnection, with beta coverage limited to equities only.
The regulatory landscape remains unresolved as AI agents transition from assisting to executing trades. Under the EU's MiFID regime, no publicly available AI tool is authorized to provide investment advice, with supervisors emphasizing transparency, auditability, and human oversight. When autonomous agents execute loss-making trades, accountability becomes genuinely ambiguous between platforms, model providers, and users. As noted by Investing.com, the competitive response involves tighter process controls rather than proven reliability, with platforms selling guardrails rather than raw AI capability. The coming period is expected to be defined less by AI's technical capabilities in markets than by regulatory requirements for human oversight and accountability before automated decision-making is deemed acceptable. Regulation is likely to become the next defining factor, ensuring that as decision-making authority migrates to machines, accountability doesn't quietly disappear along with it.
