Recent research has cast a shadow over the promising realm of AI-powered crypto trading platforms, revealing that most users are not only failing to profit but are also incurring significant losses. Conducted by researchers from Pantera Capital, Stanford University, IC3, and Ava Labs, the study examined 11 AI trading agent platforms operating on the Solana blockchain, scrutinizing over 925,000 wallets involved in these transactions.
The findings, published in their paper titled "Paper Agents, Paper Gains: An Empirical Analysis of DeFi Investment Agents," showed that users collectively lost $191.7 million, despite these same platforms holding $34.3 million in paper gains within their treasuries. This discrepancy raises urgent questions about the effectiveness and transparency of these trading agents, which had previously attracted billions in token valuations between late 2024 and early 2025.
Market Promises vs. User Reality
At their peak, the collective user gains across these platforms reached an impressive $2.4 billion. However, by February 2026, the situation had drastically changed, with users facing a substantial net loss. The study revealed that the average token price across all platforms dropped by 93% from their all-time highs, a stark contrast to Solana's own 54% peak-to-trough decline during the same period.
Only a small fraction of users benefited from their investments. The top 1% of profitable wallets, totaling just 2,590 out of 259,016 wallets that reported gains, captured an astonishing 81.4% of total profits, amounting to $1.81 billion. The most notable single profit came from the ai16z/Eliza platform, where one user netted $158.2 million. In contrast, 62.2% of participants, representing 575,246 wallets, ended up realizing losses, with median returns skewed negatively across nearly all platforms assessed.
Misleading Claims on Trading Autonomy
A critical aspect of the research highlighted that many AI platforms were not meeting their market claims. Out of the ten projects analyzed, only three were genuinely executing trades autonomously. The remaining platforms primarily provided investment advice, ran simulations, or required manual user approval for each trade.
Interviews with developers further underscored this gap. For instance, the ElizaOS team noted that "LLMs cannot trade well" without human intervention, while the Virtuals Protocol acknowledged that true autonomous trading execution remains rare, even among its extensive portfolio of over 17,000 agent launches.
The researchers did not label these platforms as fraudulent; instead, they suggested that the current market is still in its early stages, lacking the necessary infrastructure for AI agents to operate effectively.
As the AI and crypto markets continue to evolve, the findings of this study serve as a cautionary reminder. Investors drawn in by the allure of AI-driven trading must approach this complex environment critically, recognizing that the technology is still developing and may not yet meet the high expectations set by its advocates.
Looking Ahead
The implications of these findings extend beyond individual user losses. They reflect broader challenges facing the integration of AI within financial markets. As AI technologies advance and the crypto space matures, it may become possible to develop more reliable trading agents. However, until significant improvements are made, users should remain alert and informed about the capabilities and limitations of the platforms they choose to engage with.
The stories that move AI & crypto markets — before the market reacts.
Free. 7am ET. Five stories. 62,400 readers.
