
Die cooperation of the two companies Paradigm and OpenAI is a result of the realization that AI agents are increasingly able to decompile programs, analyze and manipulate the code and can thereby cause immense damage.
The OpenAI Paradigm EVM Bench is based on an extensive collection of documented vulnerabilities derived from professional audits, bug bounty competitions, and security analysis of the L1 blockchain.
The framework tests three core capabilities: identifying critical vulnerabilities, exploiting them in an isolated test environment, and then fixing the bugs through code patches.
OpenAI emphasizes that in the future, AI models will have to be evaluated in economically realistic but shielded environments in order to exclude risks for real assets. Paradigm adds that the performance of the AI models has increased significantly.
While previous generations found less than 20% of critical bugs, current models such as GPT 5.3 Codex already achieve success rates of over 70%.
More and more developers are using AI tools to create or optimize smart contracts, but not all automatically generated code is secure. AI agents are already able to automate attacks on smart contracts secured by supposedly complex defense mechanisms – with disturbing success.
The latest DeFi exploits show that even small vulnerabilities can cause millions of dollars in damage. Ethereum is therefore faced with the challenge of modernizing its security mechanisms before the AI attacks on smart contracts become even more sophisticated and successful.
The EVM Bench is the first open-to-comparison measurement tool for AI-supported security analyzes in the Ethereum system. This could be used to develop an industry standard that would help both auditors and developers identify risks early.
Image created with ChatGPT-AI (DALL E)
The EVM Bench creates additional transparency for institutional investors who are increasingly entering staking infrastructure and DeFi protocols. That’s why the cooperation between OpenAI and Paradigm is a turning point.
Ethereum must prepare for a future in which AI agents are not just passive system tools, but active system participants.
No Comments