OpenAI Launches Blockchain Vulnerability Tool
OpenAI has launched EVMbench, a new framework designed to detect and exploit vulnerabilities in blockchains. The tool signals a growing focus on the security of decentralized applications and smart contracts. This development highlights an increasing overlap between traditional penetration testing skills and the specialized field of blockchain security.
- The EVMbench framework was co-developed with the crypto investment firm Paradigm and uses a dataset of 120 real-world vulnerabilities sourced from 40 professional security audits. - It evaluates AI agents on three specific security functions: their ability to detect vulnerabilities, patch the flawed code, and successfully execute an end-to-end exploit in a controlled, sandboxed blockchain environment. - The launch of such a tool is timely, as over $1.7 billion was lost to smart contract exploits in 2024, highlighting the financial risks of bugs in the immutable code of blockchains. - Early benchmark results showed OpenAI's GPT-5.3-Codex model successfully exploiting vulnerabilities in 72.2% of test cases, demonstrating a high proficiency in offensive security tasks. - Common smart contract flaws that security professionals test for include reentrancy attacks, integer overflows, and logic errors, which allow attackers to drain funds or disrupt operations. - The global blockchain security market was valued at over $5 billion in 2025 and is projected to experience massive growth, with some estimates reaching over $495 billion by 2034, driving demand for specialized security talent. - In response to the growing risks, OpenAI has also committed $10 million in API credits through its Cybersecurity Grant Program to support defensive research and help secure critical open-source projects.
