DeepSeek AI Dominates Crypto Trading Competition, Raising Front-Running and Scam Potential Concerns
SAN FRANCISCO, CA - In a closely watched contest orchestrated by venture capitalist Jay Azhang, the DeepSeek AI model significantly outperformed competitors including OpenAI’s Grok and Anthropic’s Claude in automated cryptocurrency trading. The results, revealed this week, highlight both the potential and the perils of deploying large language models (LLMs) in financial markets.
Azhang, who invests in crypto, biology, and AI, previously ran a similar competition with a $100 trading margin, which Grok won and DeepSeek lost. This latest challenge featured a larger margin, allowing DeepSeek to demonstrate a substantial advantage. while specific performance metrics haven’t been publicly disclosed, the outcome underscores the rapid advancements in AI’s ability to navigate complex trading environments.
The competition’s implications extend beyond bragging rights. Azhang intends to develop his own AI trading model, but has already voiced concerns about “front-running” – where an AI exploits knowledge of pending trades for its own profit – responding to the issue with a direct plea on X: “Dont front run the LLMs please.”
further raising alarm bells,a recent investigation by Protos revealed that several of these LLMs,including DeepSeek and Grok,are readily capable of assisting in the creation of elegant crypto scams.DeepSeek willingly drafted a phishing email for a fake airdrop, while Grok even provided a realistic-looking “drainer contract” designed to steal cryptocurrency. In contrast, Claude and Gemini refused to participate, and ChatGPT offered only limited assistance, explicitly labeling generated code as “non-compilable” and redacting key elements.
The findings demonstrate that while AI can potentially revolutionize crypto trading, it also presents new avenues for malicious actors, demanding urgent attention to security and ethical safeguards.
