AI-Driven DeFi Hack “Epidemic” Fears May Be Early—Watch Next Wave

Crypto security concerns peaked in April after a string of high-profile hacks raised the question of whether “AI-driven” exploit discovery had changed the threat landscape. The debate resurfaced in May, when OpenZeppelin founder Manuel Aráoz said, in response to losses from exploits, that “all of DeFi unsafe” — a warning that many interpreted as evidence the industry’s defenses may be falling behind.

Yet as months passed, the pattern appeared less apocalyptic. Dragonfly managing partner Haseeb Qureshi argued that fears of a DeFi “hackpocalypse” looked like a “false alarm,” pointing to a lower year-to-date rate of hacked dollars per month and a declining median hack size. The real question now is not whether AI can assist attacks — but how quickly it is changing where losses come from, and what remains unchanged.

Key takeaways

CertiK reports that more than $1.3 billion was lost across 344 security incidents in the first half of 2026, though confirming whether AI was involved in each exploit can be difficult. Security researchers say AI is currently best understood as a “scale” multiplier that makes existing attack paths cheaper and faster, not necessarily as the replacement for human mistakes or weak controls. Chainalysis data suggests AI-enabled scams are far more profitable than traditional scams and highlights a surge in impersonation fraud using deepfakes and face-swapping. Earlier attack dynamics still dominate major losses: wallet compromise and compromised keys/signers/infrastructure account for a large share of theft, indicating fundamentals remain the key risk driver.

The DeFi “hackpocalypse” debate: data says slower, not safe

The panic cycle began when April’s exploit losses stood out for their size. In May, Aráoz drew attention to the broader implication: if attackers can automate vulnerability discovery and exploit development, DeFi risk could appear systemic rather than protocol-specific. That framing quickly fed fears that agentic AI would turn smart-contract failures into a cascading wave of failures across the ecosystem.

But Qureshi’s counterpoint was about timing and metrics. He argued that even with April’s major losses included, the year to date shows a lower rate of hacked dollars per month and a declining median hack size by year, suggesting the industry may not be entering a new era of constant catastrophic compromise.

Hacken’s Stephen Ajayi attempted to reconcile both narratives. He said the “hackpocalypse” story is overstated if it implies AI has already displaced traditional causes such as compromised keys, weak infrastructure, and human error. At the same time, he warned that “not dominant yet” does not mean “not coming,” describing the current stage as a gap between hype and incident data that he expects to narrow as attacker capabilities improve.

What defenders can and can’t prove about AI in attacks

When it comes to attributing individual hacks to AI assistance, investigators face a hard constraint: evidence is often indirect. CertiK’s H1 2026 report tracked 344 security incidents totaling more than $1.3 billion in losses, but CertiK senior blockchain investigator Natalie Newson said it can be difficult to “prove whether AI was used to find an exploit.”

Instead of direct attribution, Newson said analysts look for circumstantial indicators. One pattern she highlighted is an increase in exploits targeting older smart contracts and contracts whose source code had not been verified. CertiK’s findings also show that many vulnerable code instances took time to be abused: in the first half of 2026, 73 code vulnerability incidents were deployed for at least a year before being exploited, compared with 45 for all of 2025.

Those numbers point toward an operational shift: AI may help attackers scan and analyze much larger volumes of code than would be practical manually, then prioritize weaknesses with higher likelihood of exploitation. Newson framed the change as optimization rather than novelty — AI can help examine codebases, detect patterns tied to known vulnerabilities, summarize complex logic, and direct attention to “areas for deeper review.”

In practical terms, even if AI is not inventing new exploit categories, it can lower the friction required to find and pursue them. That matters for DeFi and other on-chain ecosystems because older contracts and unverified codebases have historically been harder to review comprehensively, creating an enduring risk surface that automated analysis can widen.

AI’s strongest impact may be profitability and fraud at scale

Chainalysis describes AI’s biggest effect not only in technical exploits, but in how criminal operations scale across victims. Sully Hanif, head of UK public sector at Chainalysis, said the firm’s 2026 crypto crime reporting found AI-enabled scams are 4.5x more profitable than traditional scams — extracting $3.2 million per operation versus $719,000.

“AI is enabling scammers to reach and manipulate far more victims simultaneously.”

Chainalysis also points to impersonation fraud as a major pressure point. In 2025, impersonation scams increased more than 1,400% year over year, with criminals using AI-generated deepfakes and face-swapping software available through Telegram marketplace ecosystems. Hanif argued that attackers have “supercharged existing playbooks,” while fraud-as-a-service offerings provide modular, turnkey components that AI further enhances.

Beyond impersonation, Chainalysis has also focused on the risks posed by unverified smart contract source code. The firm identified $36.7 million stolen from protocols where smart contract source code had never been publicly verified. Hanif warned that attackers can use large language models to reverse engineer bytecode and identify vulnerabilities at scale, reducing the research effort required before exploitation.

This is where the threat and the fundamentals connect. Even if the underlying vulnerability classes are familiar, the ability to industrialize reconnaissance and victim targeting can dramatically change outcomes. Newson similarly noted that AI’s most significant risk may appear where human effort has traditionally been the bottleneck, such as impersonating support staff, organizing video-based social engineering, or tailoring messaging to individuals. She added that attackers may increasingly not need advanced technical expertise or strong language skills, broadening the pool of potential adversaries.

Where the biggest losses come from — and what AI does not yet change

Even with these shifts, the largest losses reported for the period may still be carried out without AI as the deciding factor. CertiK’s H1 2026 report found wallet compromise remained the most damaging attack vector, accounting for more than $444 million in losses across just 33 incidents during the first half of the year.

Hacken’s Q2 2026 Web3 security report reinforced that pattern from a different angle. It found that roughly 88% of value stolen in the second quarter was tied to compromised keys, signers, and operational infrastructure rather than smart contract bugs. The report attributed much of that share to two North Korea-linked incidents involving Drift Protocol and KelpDAO.

Taken together, these findings suggest AI is currently acting as an amplifier rather than a total replacement. Ajayi’s assessment aligned with that view: AI may help identify the people and processes most vulnerable, generate more convincing phishing campaigns, parse public code, and accelerate exploit development — but whether those attacks succeed, and how large the resulting losses become, still hinges on compromised governance, operational security gaps, and weak infrastructure.

The defensive side is also moving toward prevention

AI is not only a tool for criminals. Chainalysis and CertiK narratives both acknowledge that security teams are beginning to shift from purely reactive incident response toward earlier intervention. Hanif said investigators are moving from “reactive to preventative,” and that tools exist now to stop scams before victims lose money.

Newson’s view was more balanced: AI is likely to enhance both attack and defense capabilities, with outcomes depending on which side can integrate and operationalize the technology most effectively. In other words, the advantage may not be permanent for either side — it may be earned through deployment discipline, faster review pipelines, stronger identity and access controls, and better verification practices.

For investors, protocol teams, and security practitioners, the next signal to watch is whether the proportion of losses shifts away from key compromise and operational failures toward vulnerabilities that are more directly discoverable at scale — particularly in older and unverified code. If that change accelerates, the “hackpocalypse” fears may gain new statistical support; if not, the data may continue to show that AI changes the pace, but not the core failure modes.

This article was originally published as AI-Driven DeFi Hack “Epidemic” Fears May Be Early—Watch Next Wave on Crypto Breaking News – your trusted source for crypto news, Bitcoin news, and blockchain updates.