OpenAI has launched a recent benchmark that assesses how well various AI models detect, patch, and even exploit vulnerabilities discovered in crypto sharp contracts.
OpenAI released on Wednesday, working with investment firm Paradigm and cryptocurrency security firm OtterSec to assess how much artificial intelligence agents could theoretically exploit of 120 sharp contract vulnerabilities.
Anthropic’s Claude Opus 4.6 won with an average “detection reward” of $37,824, followed by OpenAI’s OC-GPT-5.2 and Google’s Gemini 3 Pro with $31,623 and $25,112, respectively.
As AI agents become more and more productive at performing basic tasks, OpenAI he said it is becoming increasingly vital to evaluate their performance in “economically significant environments.”
“Smart contracts secure billions of dollars in assets, and AI agents are likely to be transformative for both attackers and defenders.”
“We expect to see an increase in agent payments in stablecoins and help establish them in a domain of new practical importance,” OpenAI added.
Circle CEO Jeremy Allaire predicted on January 22 that within five years, billions of AI agents will be transacting using stablecoins for everyday payments on behalf of users, while former Binance CEO Changpeng “CZ” Zhao also recently tipped that cryptocurrencies will become the “native currency of AI agents.”
The need to test AI agent performance to detect vulnerabilities comes as attackers stole $3.4 billion in crypto assets in 2025, a marginal augment from 2024.
Related: China’s AI leader will shape the future of cryptocurrencies
EVMbench exploited 120 selected vulnerabilities from 40 sharp contract audits, with most of them coming from open source auditing competitions. OpenAI said it hopes the benchmark will assist track AI’s progress in detecting and mitigating sharp contract vulnerabilities at scale.
Smart contracts weren’t made for humans: Dragonfly
Dragonfly managing partner Haseeb Qureshi wrote in a post to X on Wednesday he said Cryptocurrency’s promise to replace property rights and legal contracts has never materialized, not because the technology failed, but because it was never designed with human intuition in mind.
Qureshi said signing vast deals still seems “scary,” especially since wallet drains and other risks are always present, while bank transfers rarely pose the same concern.
Instead, Qureshi believes that the future of cryptocurrency transactions will be facilitated by autonomous AI-powered wallets that will take care of these threats and manage intricate operations on behalf of users:
“Technology often gets implemented when a complement finally comes along. GPS had to wait for the smartphone, TCP/IP had to wait for the browser. In the case of cryptocurrencies, we could simply find it in AI agents.”
Warehouse: IronClaw competes with OpenClaw, Olas launches bots for Polymarket – AI Eye
