Morning Minute: OpenAI and Paradigm Turn Focus to Smart Contracts

Core Insights - OpenAI and Paradigm are collaborating to enhance smart contract security through the launch of EVMbench, an open benchmarking framework aimed at evaluating AI agents in detecting vulnerabilities [2][3] - The initiative comes in response to the significant financial losses from smart contract exploits, which have exceeded $5 billion in the past two years [7] Group 1: Project Overview - EVMbench evaluates AI agents across three modes and utilizes 120 high-severity vulnerabilities from 40 real-world audits, primarily sourced from Code4rena contests [2] - The project aims to improve the ability of AI models to detect critical bugs, which have seen an increase in exploitability from under 20% to over 70% [3] Group 2: Financial Commitment and Future Outlook - OpenAI has expanded its private beta of Aardvark, a dedicated security research agent, and has allocated $10 million in API credits through its Cybersecurity Grant Program to bolster defensive research in the crypto space [3] - Paradigm anticipates that a growing portion of future audits will be conducted by AI agents, positioning EVMbench as both a preview and an accelerant for this shift [3] Group 3: Implications and Risks - The dual-use nature of EVMbench is highlighted, as the same AI capabilities that enhance security can also be exploited for malicious purposes, creating a potential hacking playbook if misused [7]