Blockchain Security at Risk: AI Shows Risks in Post-2025 Smart Contracts
According to Anthropic, an AI research firm, sophisticated AI agents have found flaws in new smart contracts. These flaws may have resulted in millions of dollars’ worth of exploits, underscoring the rising danger AI poses to blockchain security. According to the company’s most recent report, it tested several models in a simulated blockchain environment, including Claude Opus 4.5 and Claude Sonnet 4.5. Smart contracts deployed after March 2025 that had previously been exploited were the subject of the testing. According to Anthropic, the AI agents stole $4.5 million in simulated funds by effectively exploiting 17 out of 34 test contracts.
Anthropic’s Study Highlights Rising AI Threats to Blockchain Security
207 contracts were effectively exploited by AI models out of its benchmark of 405 contracts that were implemented across Ethereum, BNB Smart Chain, and Base between 2020 and 2025. The research said that the mock revenue from these exploits was $550 million. The mission assigned to Sonnet 4.5 and GPT-5 by Anthropic was to scan 2,849 recently deployed contracts that were known to be vulnerability-free. They did find two new zero-day vulnerabilities, however, that could be exploited for $3,694. Unprotected read-only functions and authorization issues that permitted user fund withdrawals were among the vulnerabilities that might allow AI agents to alter token supplies. In the fee withdrawal logic, there were also missing validations.
More than half of the blockchain exploits carried out in 2025 — presumably by skilled human attackers — could have been executed autonomously by current AI agents. Our proof-of-concept agent’s further discovery of two novel zero-day vulnerabilities shows that these benchmark results are not just a retrospective — profitable autonomous exploitation can happen today.
the report
Anthropic: AI Can Both Hack and Protect Smart Contracts
Anthropic noted that throughout the previous year, exploit revenue from stolen simulated funds increased every 1.3 months due to the rapid progress of AI.
As costs continue to fall, attackers will deploy more AI agents to probe any code that is along the path to valuable assets, no matter how obscure: a forgotten authentication library, an obscure logging service, or a deprecated API endpoint,
the report
Anthropic, however, highlighted the advantages of AI technology. They pointed out that vulnerabilities might be patched using the same agents. The smart contracts exploitation benchmark (SCONE-bench) dataset will be made publicly available, according to the company. The goal of this project is to assist developers with contract testing and patching.
For more up-to-date crypto news, you can follow Crypto Data Space.
