OpenZeppelin founder warns all of DeFi is unsafe amid security breaches

Warnings about decentralized finance are nothing new. Smart contract exploits, bridge failures, governance attacks and phishing campaigns have accompanied the sector since its earliest years. Investors have largely accepted that higher returns often come with higher technical risk. What makes Manuel Aráoz’s latest warning different is not simply its severity, but the reason behind it.

The OpenZeppelin co-founder, whose company helped establish many of the security standards used across Ethereum and DeFi, said in May that he now considers “all of DeFi” unsafe and has privately advised friends and family to exit even large, established protocols such as Aave, MakerDAO and Compound. His argument is not that developers have become less careful. It is that artificial intelligence has fundamentally altered the economics of software security. According to Aráoz, coding agents have become “superhuman” at finding vulnerabilities, while defenders still face the impossible task of eliminating every critical flaw before attackers find one.

That deserves attention because it challenges one of the assumptions on which DeFi has been built: that security improves steadily as protocols mature.

The Security Equation Has Changed

Traditional software security has always been asymmetric. Developers must identify and fix every significant vulnerability before release. An attacker only needs one weakness.

Until recently, finding those weaknesses required highly specialised expertise, time and resources. AI changes that calculation.

Modern coding models can analyse large codebases, compare implementations, identify unusual logic paths and generate exploit hypotheses far faster than a human researcher working alone. While they are not perfect, they reduce the cost of searching for vulnerabilities.

For DeFi, this is particularly uncomfortable because smart contracts are public by design. Every protocol publishes the code that secures billions of dollars in assets. Transparency has always been presented as one of blockchain’s strengths because anyone can inspect the contracts. But transparency works both ways. The same openness that allows auditors to review code also gives increasingly capable AI systems unlimited access to analyse it.

That is the asymmetry Aráoz believes has become much harder to overcome.

Audits Are No Longer Enough

For years, the industry has relied heavily on audits as its primary trust signal.

A protocol would announce funding, publish an audit report, launch a bug bounty programme and reassure users that multiple security firms had reviewed the contracts.

Those measures still matter. They remove obvious mistakes and improve code quality. But they should not be confused with guarantees.

An audit captures the state of the code at a particular moment. It cannot predict every interaction with future protocols, every governance decision, every market condition or every creative attack vector that may emerge months later.

The problem becomes even harder when protocols are deeply interconnected. Modern DeFi applications rarely exist in isolation. Lending protocols connect to liquid staking systems, which interact with bridges, oracles, liquidity pools, cross-chain messaging and governance frameworks. Each additional dependency expands the attack surface.

Security therefore becomes an ecosystem property rather than a property of one protocol.

Complexity Has Become The Real Enemy

DeFi has succeeded because it is composable. Developers can combine existing protocols to build new financial products without asking permission.

That composability has accelerated innovation. It has also increased systemic complexity.

Every integration introduces assumptions. Every oracle depends on external data. Every bridge depends on another security model. Every governance upgrade creates new operational risk. Every liquidity layer creates another potential point of failure.

The result is that many modern exploits do not originate from simple programming mistakes. They emerge from unexpected interactions between otherwise well-designed systems.

Academic research on DeFi attacks has reached similar conclusions. Reviews of hundreds of real-world incidents show that vulnerabilities extend well beyond coding errors to include oracle manipulation, governance weaknesses, permissionless interactions and economic design flaws.

This is why simply demanding “more audits” is becoming an incomplete answer.

AI May Benefit Attackers First

Every technological advance creates opportunities for both defenders and attackers.

Security companies are already using AI to improve code review, automate testing and assist vulnerability research. Protocol developers are increasingly experimenting with AI-assisted verification and monitoring.

But attackers often benefit first because they operate under fewer constraints.

A legitimate security review must minimise false positives, document findings, coordinate disclosure and avoid disrupting production systems. Attackers have no such obligations. They can run thousands of automated analyses, discard failed attempts and immediately exploit successful discoveries.

The economics therefore favour offence.

This does not mean AI inevitably destroys DeFi security. It means defensive security needs to improve faster than it has historically.

Trust Will Depend On Operations, Not Marketing

If Aráoz’s assessment proves directionally correct, DeFi projects will need to compete differently.

Marketing centred on audits, TVL and token incentives will become less persuasive than operational resilience.

Institutional investors and sophisticated users are likely to ask different questions.

How quickly can vulnerabilities be identified?

Can contracts be paused safely?

How are governance upgrades controlled?

What monitoring exists after deployment?

How are dependencies mapped?

How much value can be exposed before emergency mechanisms activate?

How often are threat models updated?

Those questions focus less on the existence of risk and more on the ability to manage it.

That shift mirrors developments in cybersecurity generally. Organisations increasingly accept that breaches may occur. Competitive advantage comes from detection, containment and recovery rather than pretending perfect prevention is possible.

Does This Mean DeFi Is Finished?

Probably not.

History suggests that major technologies often become safer through repeated failures rather than before them. The internet, cloud computing and online banking all experienced periods where attackers consistently outpaced defenders before security standards matured.

The difference is that financial losses in DeFi are immediate, public and usually irreversible.

That creates much stronger pressure to improve security architecture.

AI may ultimately strengthen DeFi if it becomes a more effective defensive tool than offensive one. Automated code verification, continuous monitoring, formal verification assisted by AI and real-time anomaly detection all have the potential to improve protocol resilience.

The question is whether those improvements can arrive quickly enough.

What Investors Should Take Away

Aráoz’s warning should not be interpreted as a call to abandon every blockchain application.

It should be understood as a reminder that “blue-chip” status is not the same as low risk.

Protocol size does not eliminate smart contract exposure.

Past audits do not eliminate future vulnerabilities.

Reputation does not eliminate operational mistakes.

Yield always has a corresponding source of risk.

Investors therefore need to evaluate security with the same seriousness they apply to returns.

Questions about governance, treasury management, upgrade processes, insurance arrangements, dependency mapping and incident response deserve at least as much attention as annual percentage yields.

The Next Competitive Advantage

For several years, DeFi projects competed primarily on innovation.

Then they competed on liquidity. Today they increasingly compete on security. The next stage may be continuous security rather than one-time security.

Protocols that can demonstrate AI-assisted monitoring, transparent operational controls, rapid response capabilities, independent verification and conservative exposure management are likely to attract more long-term capital than those relying mainly on incentives. Aráoz’s statement is deliberately provocative, but it raises a serious strategic question. If AI allows attackers to discover vulnerabilities at machine speed, can DeFi continue relying on security practices designed for human speed? That question is likely to shape the next phase of decentralized finance far more than the next token launch or yield opportunity.