AI Agents Expose Blockchain Vulnerabilities in Exploits Worth Millions, Says Anthropic

• AI agents exploit blockchain vulnerabilities, simulating theft of millions in funds.
• Anthropic reveals AI’s potential to both attack and defend blockchain contracts.
• AI-driven security measures crucial to safeguarding smart contracts from vulnerabilities.

According to Anthropic’s latest report, AI models have proven their capability to uncover vulnerabilities in blockchain smart contracts, exposing the growing risks they pose to digital assets. The company tested advanced models, including Claude Opus 4.5 and Claude Sonnet 4.5, in a mock blockchain environment. In this experiment, the AI agents identified and exploited flaws in previously exploited smart contracts, ultimately simulating the theft of $4.5 million from 17 of the 34 tested contracts.

The benchmark study, which analyzed 405 contracts deployed between 2020 and 2025 across major blockchain networks like Ethereum, BNB Smart Chain, and Base, highlighted a staggering success rate for AI agents. These agents managed to exploit 207 contracts and siphon off an impressive $550 million in simulated funds. Furthermore, the study revealed that AI models like Sonnet 4.5 and GPT-5, when tasked with scanning nearly 3,000 recently deployed contracts, uncovered two novel zero-day vulnerabilities, potentially leading to exploits valued at $3,694.

The increasing use of AI agents in smart contract security has raised alarm bells for the blockchain community. Anthropic’s report notes that the rapidly advancing AI technology could become a significant tool for both malicious actors and defenders. While skilled human attackers might exploit vulnerabilities autonomously, the same AI models could be deployed to identify and patch these weaknesses.

Also Read: XRP Whale Wallets Drop Sharply, But There is a 48,000,000,000 Coin Twist

Rising Threat and the Need for AI-Driven Defense

The report emphasizes the growing prevalence of AI in identifying vulnerabilities in blockchain contracts, as exploit revenues doubled every 1.3 months over the past year. The continuous improvement in AI’s capabilities suggests that attackers will increasingly deploy agents to probe various code vulnerabilities—ranging from obscure authentication libraries to deprecated API endpoints.

However, Anthropic sees potential in using AI for positive purposes as well. The company plans to open-source its smart contract exploitation benchmark (SCONE-bench) dataset to empower developers. This will allow them to test and fix smart contracts before they are deployed, strengthening blockchain security and reducing the risk of future exploits.

The rapid advancements in AI technology present both an opportunity and a challenge for blockchain developers. As AI agents continue to evolve, the urgency to integrate AI-driven defense mechanisms has never been more critical. Anthropic urges developers to adapt their security strategies and adopt AI tools to safeguard against emerging threats.

Also Read: BitMine’s Ethereum Holdings Surge Despite Falling Crypto Prices

The post AI Agents Expose Blockchain Vulnerabilities in Exploits Worth Millions, Says Anthropic appeared first on 36Crypto.