Recent research reveals that AI agents are increasingly capable of autonomously exploiting vulnerabilities in blockchain smart contracts, potentially leading to significant financial losses. A new benchmark study showcases the economic impact of these attacks and highlights the urgent need for proactive defensive strategies against advanced AI-powered cyber threats. The study found that leading AI models like Claude Opus 4.5 and GPT-5 generated over $4.6 million in exploit revenue by exploiting vulnerabilities in contracts deployed after their knowledge cutoff in March 2025. This finding underscores a worrying trend: AI’s growing potential for offensive operations within blockchain ecosystems, further driving the need for robust defensive measures.
The study, which utilized the SCONE-bench, analyzed 405 real-world contracts exploited between 2020 and 2025. It discovered two novel zero-day vulnerabilities in newly deployed contracts, successfully generating gains of $3,694. The findings highlight the autonomous ability of AI agents to discover and exploit previously unknown vulnerabilities, demonstrating the expanding cyber risk posed by sophisticated AI models.
This research also provides insights into the financial impact of these exploits, showcasing a direct correlation between AI model efficiency, vulnerability exploitation, and attack profitability. The study’s focus on monetary outcomes adds crucial context for security stakeholders who prioritize quantifiable data in assessing risks.
Moreover, the study explores the dual-use nature of AI for both defense and offense within the blockchain space. This calls for proactive implementation of AI-powered defense systems to counter increasingly sophisticated cyber threats. By utilizing this technology, developers can proactively patch vulnerabilities before malicious actors exploit them. The SCONE-bench serves as a valuable resource for accelerating defensive efforts against AI-driven attacks.
Meanwhile, Cocoon, a confidential-compute network built by a Telegram co-founder on The Open Network ($TON), provides an alternative approach to safeguarding user data during AI computations.
The threat posed by AI agents demands a proactive and anticipatory approach from the blockchain community. By investing in robust defense mechanisms and adopting best practices for secure development, we can mitigate the risks of unchecked AI exploitation and protect the integrity of the blockchain ecosystem.