Claude, the computer development-focused AI tool from Anthropic, released a major update last week that impacts the crypto industry.
Developers are now coding not only with full desktop access using Claude’s brand new CoWork tool, but are now using it to vibe audit smart contracts.
Smart contracts are blockchain-based computer programs that automatically carry out the terms of an agreement when technical conditions are met.
They have operated for over a decade and move around billions of dollars worth of digital assets every day but they’re are still prone to hacks.
Vibe auditing, like vibe coding, relinquishes important labor to AI tools. If a developer doesn’t understand how to fully audit financial smart contracts, the AI industry has a plugin for that.
In today’s new age of crypto security, developers can simply activate so-called “skills” in Claude to outsource auditing work to an assortment of cybersecurity tools. For example, one Claude skill will ostensibly check specification-to-code compliance with blockchain audits.
“Use this skill when you need to audit smart contracts against whitepapers or design documents,” says a Trail of Bits whitehat hacker about their spec-to-code compliance skill.
Another Claude skill from the Trail of Bits marketplace allows vibe auditors to tell Claude to check for variants of smart contract bugs using “pattern-based analysis.”
If the AI doesn’t find a variant, that obviously doesn’t guarantee the absence of a bug. Disclosure, of course, is up to the vibe auditor.
Read more: Anthropic gamble could save FTX customers from total ruin
A variety of Claude code smart contract vibe auditing skills are available via the Claude Code marketplace.
Whether a client is looking to move 10 or 10 billion dollars using a smart contract, vibe auditors can check for footgun designs, compiler-induced timing side-channels, sharp edges, state-changing entry points, property-based tests, git history, and plenty of other cybersecurity jargon.
To its credit, Trail of Bits is at least offering these tools to smart contract developers to help them begin the process of preventing bugs, rather than withholding all tools from Claude.
The cybersecurity company has also licensed its Claude skills under Creative Commons, and includes plenty of warnings and disclosures in its documentation that AI security checks are incomplete and prone to errors.
It repeatedly recommends manual reviews of all code by a team of cybersecurity experts prior to deploying a smart contract on a public blockchain.
