The long-standing disconnect between investment analysis and execution is rapidly dissolving, as new platforms enable direct trading within artificial intelligence (AI) chatbots. For investors accustomed to researching opportunities with tools like ChatGPT or Claude only to switch to a separate brokerage application for execution, this represents a significant operational shift. Liquid’s new Co-Invest app, embedded directly within ChatGPT and Claude, now allows users to fund accounts, analyze markets, and place live trades without ever leaving the AI conversation.
Integrating Analysis and Execution
Launched on Tuesday (May 26), Liquid’s Co-Invest app aims to streamline the investment process by consolidating the entire workflow into a single interface. The platform supports a broad array of markets, covering more than 500 instruments including crypto, equities, foreign exchange, Polymarket prediction market positions, and pre-IPO secondaries. Users gain the ability to set crucial risk management parameters, such as stop-loss and take-profit levels, directly within the chat environment, with each order requiring a tap for confirmation.
A key feature emphasized by Liquid is its non-custodial operation, ensuring that user funds remain in their own wallets, rather than being held by the platform. Franklyn Wang, Liquid’s CEO, articulated the strategic rationale behind this integration to The Block, stating, “Millions of people already listen to AI to figure out what to invest in.” He added that putting execution in the same interface is simply “the next logical step” in the evolution of AI-driven financial tools.
Accelerating Adoption and Market Momentum
The market Co-Invest is entering is already demonstrating substantial growth and user engagement. Data from PYMNTS Intelligence, as of December, revealed that 37% of power AI users reported utilizing native AI platforms as their primary tool for managing finances and banking. This trend is not confined to early adopters; among mainstream users, the share of those adopting AI for financial tasks doubled in a single month. However, the report also highlighted a significant comfort gap, with only 14% of light users feeling comfortable employing AI for financial tasks, indicating that current adoption remains concentrated among more engaged users.
Liquid’s Co-Invest is not an isolated development but rather part of a broader industry movement towards integrating AI with transactional capabilities. Several other prominent financial technology firms have made similar strategic moves:
- Robinhood introduced Agentic Trading and an Agentic Credit Card, enabling AI agents to execute trades and card payments on behalf of customers via a dedicated virtual card with user-defined spending limits.
- MoonPay recently launched its own ChatGPT integration specifically for crypto purchases.
- OpenAI, in May, rolled out personal finance tools for ChatGPT Pro users through a Plaid integration.
- Gemini unveiled agentic trading functionalities, allowing AI models to connect directly to exchange accounts.
The consistent pattern across these initiatives is a clear shift: AI assistants are evolving from merely providing an information layer to actively participating in the transaction layer, fundamentally altering how users interact with financial markets.
Navigating the Risks of Automated Execution
While the convenience argument for direct AI-driven trading is compelling, the inherent risks are more complex and difficult to dismiss. Regulators are already signaling concerns. FINRA’s 2026 Annual Regulatory Oversight Report specifically flagged “hallucinations” as a compliance concern for broker-dealers. The regulator warned firms to develop robust procedures for AI agents that might operate beyond a user’s intended scope, defining hallucinations as instances where a model generates inaccurate or misleading information but presents it as factual.
In financial contexts, the implications of such errors are direct and potentially severe. Fortune reported in April that large language models are inherently stochastic, meaning that even well-tuned agents can produce errors with the same confidence as correct outputs. When an AI agent is positioned directly atop a brokerage account, a hallucinated price, a misread regulatory filing, or a misinterpreted order parameter could lead to financial transactions before a user can intervene or even notice the discrepancy.
Implementing Guardrails and User Control
Recognizing these substantial risks, both Liquid and Robinhood have incorporated specific confirmation steps and guardrails into their products. Liquid’s Co-Invest app mandates a user tap to confirm every trade, providing a crucial human-in-the-loop control point. Robinhood offers customers the flexibility to choose whether to require manual approvals for AI-executed actions. Furthermore, Robinhood restricts its agents by default to a dedicated virtual card, ensuring no direct access to primary account information and allowing users to set a spending limit.
Abhishek Fatehpuria, Robinhood’s Vice President of Product Management, stated that this design directly reflects customer feedback: the desire to empower agents with Robinhood’s capabilities, but in a secure manner. However, a fundamental question remains unanswered: whether these current guardrails are truly sufficient to protect users when following AI-generated analysis that, due to the stochastic nature of large language models, may be confidently incorrect.
As AI continues its inexorable move from analytical support to direct transactional execution in financial markets, the balance between unprecedented convenience and the imperative of robust risk mitigation will define the next chapter for investors and regulators alike. The efficacy of current safeguards in preventing confidently erroneous AI outputs from translating into real-world financial losses will be a critical determinant of widespread adoption and trust in this evolving paradigm.
