AI is not a risk of the future. It is already an amplifier, scaling up vulnerabilities in today’s blockchains.
Smart contracts, wallet firmware, oracles, governance contracts, and legacy codebases—all can be repeatedly scanned, dissected, and replicated by AI if they’re open source.
“Passed an audit” no longer means “secure.” The real danger is that vulnerabilities are being discovered faster than they can be fixed.
According to Binance Research, AI’s success rate in smart contract exploit testing is roughly twice its detection rate, and the cost of scanning a single contract has fallen to about $1.22.
Anthropic’s simulated research also shows that AI can generate working exploit scripts for real smart contracts in a sandbox and find zero-day vulnerabilities in contracts it has never encountered before.
In the first half of 2026, losses from on-chain security incidents totaled about $1.316 billion, including 204 code vulnerability incidents—a very high frequency.
The Coldcard incident illustrates the problem: a long-standing firmware flaw could allow attackers to systematically derive wallet seeds, resulting in losses ranging from tens of millions to hundreds of millions of dollars.
But let’s be clear: we can’t simply claim that all thefts happen “because of AI alone.”
AI has indeed lowered the barrier to attack and increased the likelihood that old code will be dug up. But many incidents still stem from engineering mistakes, misconfigurations, weak access controls, and inadequate audit coverage.
A more accurate assessment is this: AI is making “old vulnerabilities” lethal again.
Smart contracts, wallet firmware, oracles, governance contracts, and legacy codebases—all can be repeatedly scanned, dissected, and replicated by AI if they’re open source.
“Passed an audit” no longer means “secure.” The real danger is that vulnerabilities are being discovered faster than they can be fixed.
According to Binance Research, AI’s success rate in smart contract exploit testing is roughly twice its detection rate, and the cost of scanning a single contract has fallen to about $1.22.
Anthropic’s simulated research also shows that AI can generate working exploit scripts for real smart contracts in a sandbox and find zero-day vulnerabilities in contracts it has never encountered before.
In the first half of 2026, losses from on-chain security incidents totaled about $1.316 billion, including 204 code vulnerability incidents—a very high frequency.
The Coldcard incident illustrates the problem: a long-standing firmware flaw could allow attackers to systematically derive wallet seeds, resulting in losses ranging from tens of millions to hundreds of millions of dollars.
But let’s be clear: we can’t simply claim that all thefts happen “because of AI alone.”
AI has indeed lowered the barrier to attack and increased the likelihood that old code will be dug up. But many incidents still stem from engineering mistakes, misconfigurations, weak access controls, and inadequate audit coverage.
A more accurate assessment is this: AI is making “old vulnerabilities” lethal again.