OpenAI's workspace agents handle multi-hour business tasks; its partnership with Paradigm brings AI to smart contract security via EVMbench.
A quiet partnership between OpenAI and crypto venture firm Paradigm aims to close one of blockchain's costliest gaps: smart contract vulnerabilities. The tool, called EVMbench, uses AI agents to test and secure smart contracts on Ethereum Virtual Machine-compatible chains. It benchmarks how effectively those agents identify flaws before attackers do.
The partnership, announced February 18, 2026, lands as OpenAI rolls out workspace agents built on its Codex engine. Those agents, launched April 22, 2026, can run autonomously for hours, writing code, generating reports, and managing workflows. Codex had reached 3 million weekly active users by early 2026, with its APIs processing over 15 billion tokens every minute. OpenAI reported a 137-fold growth in non-developer users of Codex agents since August 2025. Top users now log over 60 hours of agent turns daily, often deploying multiple agents in parallel.
That kind of autonomous capability is exactly what EVMbench wants to harness for security. Smart contract exploits have drained hundreds of millions from DeFi protocols. Traditional audits are manual and slow. Automated testing exists but often misses edge cases. EVMbench aims to push AI agents to systematically probe for reentrancy attacks, oracle manipulation, and logic flaws across EVM-compatible chains.
OpenAI's earlier work in AI for code safety included the original ChatGPT agent released July 17, 2025, which could perform tasks like calendar analysis and slide deck creation. The Codex agents are a step change in duration and complexity. Applying that same engine to smart contract security is a natural extension. Paradigm, known for its deep blockchain engineering, provides the domain expertise to shape the benchmarks.
The practical question for crypto traders and developers is whether EVMbench can move beyond benchmarking into production-grade auditing. If AI agents can routinely find vulnerabilities that human auditors miss, the cost and speed of smart contract security could shift dramatically. No timeline for a public release has been announced.
Prepared with AlphaScala research tooling and grounded in primary market data: live prices, fundamentals, SEC filings, hedge-fund holdings, and insider activity. Each story is checked against AlphaScala publishing rules before release. Educational coverage, not personalized advice.