AI agents review smart contracts to identify and fix security issues that lead to crypto losses

Developers are now using AI agents to protect smart contracts that control billions in digital assets, following crypto hackers’ theft of over $3.4 billion from blockchain platforms in 2025. 

Rather than dozens of small thefts, 2025’s losses were concentrated in a few massive breaches, with just three major incidents accounting for nearly 70 % of the total value stolen. The most notable was the Bybit exchange hack, which alone siphoned off roughly $1.4 billion — one of the largest crypto thefts ever recorded

OpenAI is working with Paradigm and OtterSec to test whether AI agents can detect vulnerabilities in real blockchain spaces using its EVMbench.

AI agents review smart contracts to identify and fix security issues that lead to crypto losses

Any error in smart contract code today will affect real money belonging to big and small investors, as these automated programs manage more than $100 billion in open-source digital assets. 

And after hackers stole over $3.4 billion from crypto platforms in 2025, developers can now see just how vulnerable the system is when attackers exploit weak code. 

Relying on human audits isn’t an option anymore because live contracts face new and evolving attacks that weren’t present during the audit process. Plus, it takes a lot of time and costs a fortune as security teams must review smart contract code before deployment.

Instead of waiting for the next manual audit cycle that may come too late to stop an attack, developers are now turning to AI agents to continuously monitor live smart contracts.

It takes AI agents less time to detect hidden code irregularities than people do, who may need days or even weeks, so frameworks like the EVMbench by OpenAI make more sense for developers.

EVMbench uses AI agents in test environments to help developers understand how smart contracts may perform under real-world pressure before the actual deployment.

The agents will first detect hidden vulnerabilities in code, fix the issue without breaking the contract’s function, and then try to exploit the weakness to drain funds if the problem persists.

According to early results, AI agents are better at exploiting vulnerabilities than safely fixing them. People are now concerned that hackers could misuse AI-powered tools to exploit weaknesses in blockchain systems more efficiently than ever. 

AI agents can also create new security risks by helping hackers identify weaknesses in blockchain systems

Machines are learning to break into weak contracts faster than ever before because current AI agent systems now succeed in exploiting more than 70% of vulnerabilities compared to earlier AI models with a less than 20% success rate.

Attackers are now moving away from manual hacking methods and toward AI agents that scan large amounts of code and test different attack paths without direct human input.

And as this trend continues, experts now say AI agents will soon be able to move funds, approve transactions, and manage financial tasks automatically on behalf of users. 

American technologist Jeremy Allaire said that billions of AI agents will soon use stablecoins to send and receive payments across blockchain networks. Founder and former CEO of Binance, Changpeng Zhao (CZ), also said crypto could become the native payment layer for AI-driven systems in the future.

All these trends and predictions make AI agents more useful to both users and attackers, as they will soon interact with contracts directly in real financial environments where actual money is at stake.

Industry leaders have even raised concerns about user safety. Managing partner at Dragonfly, Haseeb Qureshi, warned that many users still worry about sending funds to the wrong address or approving a harmful transaction by mistake through crypto transactions.

To solve this problem, Qureshi proposed that AI-operated wallets could soon interact with the blockchain without users needing to understand the complex process involved.

In this way, AI agents can assist in reducing human errors in audits and in protecting smart contracts by continuously monitoring systems. However, they can also increase the rate at which attackers discover vulnerabilities in the system, enabling exploits to scale much faster.

This creates a security issue where AI systems developed to protect decentralized finance platforms could also be the most effective at attacking them if they fall into the wrong hands.

The smartest crypto minds already read our newsletter. Want in? Join them.