How effective is AI at exploiting smart contracts?

The report shows GPT-5.3-Codex achieved a 72.2% success rate in exploit mode on EVMbench. This is roughly double its detection success rate, meaning AI is currently better at breaking systems than protecting them.

Why is AI giving hackers an advantage in crypto?

AI can scan thousands of smart contracts quickly and cheaply, identifying weaknesses at scale. At the same time, fewer developers are using AI for advanced security testing, creating a gap attackers can exploit.

How cheap are AI-powered crypto exploits now?

AI-driven exploits cost around $1.22 per contract on average. This cost is projected to fall by about 22% every two months, making large-scale attacks increasingly accessible.

What is EVMbench and why does it matter?

EVMbench is a benchmark that tests how well AI can detect and exploit smart contract vulnerabilities. It uses real-world vulnerabilities from audited contracts, making it a useful measure of practical risk.

How does this impact DeFi platforms and users?

DeFi platforms rely heavily on smart contracts to manage funds. If AI makes exploits easier and cheaper, it increases the risk of hacks and losses for users.

Are AI tools also improving crypto security?

Yes, but not at the same pace as offensive use. Over 80% of developers use AI for coding, while fewer than 40% use it for advanced security testing, slowing defensive progress.

How is AI changing crypto scams beyond smart contracts?

AI is being used to create more convincing scams, including deepfakes and impersonation attacks. These scams generate 4.5 times more revenue per case than traditional methods.

AI exploits outpace detection in crypto hacks

Q: What did Binance Research find about AI and crypto hacks?

Binance Research found that AI tools are now about twice as effective at exploiting smart contract vulnerabilities as they are at detecting them. This highlights a growing security imbalance where attackers have a clear advantage.

Written by Isaac FrancisPublished May. 1 2026Last Updated May. 1 2026

Artificial intelligence tools are now twice as effective at exploiting smart contract vulnerabilities as they are at detecting them, according to research from Binance Research.

The report found that GPT-5.3-Codex achieved a 72.2% success rate in exploit mode on EVMbench, roughly double its detection capability, highlighting a growing imbalance favouring attackers.

Researchers warned that declining attack costs, now averaging just $1.22 per contract and falling further, are making large-scale automated exploits increasingly viable.

“Whether we welcome it or not, AI is currently 2x better at exploitation than at detection,”

The report said.

The widening gap is partly driven by uneven adoption, with over 80% of developers using AI for coding but fewer than 40% deploying it for advanced security testing.

Beyond smart contracts, firms such as Chainalysis report that AI-powered scams generate 4.5 times more revenue per case, with impersonation attacks surging 1,400% in 2025.

Crypto has become a primary target, accounting for 88% of detected deepfake fraud cases, as attackers increasingly use AI to scale social engineering and exploit vulnerabilities across decentralised finance systems.