How Did a $1.12 Price Trigger a $1.78 Million Loss?
Moonwell, a decentralized lending protocol deployed on Base and Optimism, lost about $1.78 million after a pricing oracle for Coinbase Wrapped Staked ETH (cbETH) returned an incorrect value of roughly $1.12 instead of about $2,200. The mispricing allowed borrowers and liquidation bots to extract value before the issue was identified.
In a post-mortem, Moonwell said a governance proposal executed on Sunday misconfigured the cbETH oracle by relying solely on the cbETH/ETH exchange rate. As a result, the protocol reported cbETH at around $1.12. Opportunistic actors used the distorted price to borrow against misvalued collateral, leaving the system with approximately $1.78 million in bad debt.
The incident did not involve a complex exploit or cross-chain bridge vulnerability. Instead, it stemmed from a basic configuration failure in oracle logic tied to a major asset, raising concerns about review processes around governance changes and price feeds.
Investor Takeaway
Was AI-Assisted Code to Blame?
The pull requests tied to the affected contracts show multiple commits co-authored by Anthropic’s Claude Opus 4.6. Security auditor Pashov publicly linked the case to AI-assisted Solidity development after identifying the co-authorship pattern.
“The developer was using Claude to write the code, and this has led to the vulnerability,” he told Cointelegraph, explaining why he associated the flaw with AI-assisted output.
At the same time, Pashov cautioned against attributing the issue solely to artificial intelligence. He described the oracle error as the type of mistake “even a senior Solidity developer could have made,” arguing that the core failure lay in insufficient validation rather than in the use of AI tools themselves.
Initially, he believed there had been no testing or audit. He later acknowledged that Moonwell said it had conducted unit and integration tests in a separate pull request and had commissioned an audit from Halborn.
In his view, the mispricing “could have been caught with an integration test, a proper one, integrating with the blockchain,” though he declined to single out any security firm for criticism.
Small Loss, Larger Governance Questions
In dollar terms, the exploit is modest compared with DeFi’s largest failures, such as the Ronin bridge hack in March 2022, where more than $600 million was stolen. Still, the Moonwell case stands out because it combines AI co-authorship, a straightforward oracle configuration error on a widely used asset, and the presence of prior audits and testing.
The incident raises questions about governance pipelines in DeFi protocols, particularly when parameter changes are executed through proposals. Even when code is reviewed and audited, implementation details tied to pricing feeds and collateral logic can introduce risk if not tested end-to-end in realistic market conditions.
Pashov said his own firm would not overhaul its auditing process in response to this case, but added that if code appeared “vibe coded,” his team would “have a bit more wide open eyes” and expect a higher density of simple flaws, even though he noted that this specific oracle issue “was not that easy” to detect.
Investor Takeaway
What Does This Mean for AI in Smart Contract Development?
The debate around “vibe coding” gained further attention after the exploit. Fraser Edwards, co-founder and CEO of cheqd, told Cointelegraph that discussions about AI in development often blur two distinct use cases.
He said the difference lies between non-technical founders prompting AI to generate code they cannot independently assess and experienced developers using AI to accelerate refactoring, explore patterns, and expand testing inside a structured engineering process.
AI-assisted development “can be valuable, particularly at the MVP stage,” Edwards said, but “should not be treated as a shortcut to production-ready infrastructure,” especially in capital-intensive systems such as DeFi.
He argued that AI-generated smart contract code should be treated as untrusted input and subjected to strict version control, clear ownership, peer review, and robust testing, particularly around access controls, oracle logic, pricing mechanisms, and upgrade pathways.
“Ultimately, responsible AI integration comes down to governance and discipline,” Edwards said, adding that code generation and validation should be clearly separated and that any contract deployed in an adversarial environment must be assumed to carry latent risk.