Why DeFi Keeps Losing Millions to Exploits

It’s been one of the worst years on record for DeFi hacks, and we’re barely halfway through.

In the first five months of 2026, more than $840 million was lost to DeFi hacks—with April alone accounting for more than $600 million stolen, led by two of the year’s biggest attacks: the $292 million KelpDAO exploit and the $285 million Drift Protocol breach.

The losses have continued into May, with THORChain halting trading after security researchers flagged a suspected cross-chain exploit affecting more than $10 million.

TrustedVolumes, Echo Protocol, Step Finance, Truebit, Resolv Labs, Volo Protocol, Rhea Finance, Verus-Ethereum bridge, and many others round out a casualty list that reads like a stress test of every trust assumption DeFi relies on, according to DeFiLlama data.

Experts Decrypt spoke to broadly agree on the diagnosis that recent DeFi hacks are exposing structural weaknesses across bridges and admin systems, while advances in AI may be helping attackers find vulnerabilities faster.

Natalie Newson, senior blockchain investigator at Web3 security platform CertiK, told Decrypt that while April was unusually severe for crypto exploits, the broader trend remains more stable and below the peak number of incidents seen in 2023.

“April 2026 was a bad month for crypto exploits; there were only three days without an exploit in which at least $10,000 was taken,” she said.

“However, when we take a look at the wider picture, the number of incidents (excluding phishing) has arguably been fairly consistent and still lower than a peak in 2023,” Newson noted, adding how April’s severity was driven by 14 exploits exceeding $1 million in losses, second only to September 2025’s 16.

The North Korea factor

Ari Redbord, Global Head of Policy and Government Affairs at TRM Labs, told Decrypt the surge traces back to a single state actor that has gone from marginal player to defining threat in five years.

“The dominant driver is North Korea, and that campaign is getting sharper, not broader,” Redbord said, noting that North Korea-linked actors accounted for 76% of global crypto hack losses in the first four months of 2026, up from 64% in 2025 and less than 10% in 2020.

“North Korea is using not only technology to attack the space, but also sophisticated and well-planned social engineering,” he said.

The year’s largest DeFi hack so far hit KelpDAO on April 18, when attackers drained about 116,500 rsETH, worth roughly $292 million, from a cross-chain bridge.

LayerZero, whose messaging infrastructure underpinned the bridge, said in the latest postmortem report that the attack began on March 6, when a developer was socially engineered, and session keys were harvested.

The cross-chain messaging protocol said the attack was attributed by Mandiant, CrowdStrike, and independent researchers to DPRK threat actor TraderTraitor, also known as UNC4899.

The structural reason DeFi keeps absorbing the hits, Redbord added, comes down to where the money sits and how it moves.

“DeFi’s cross-chain complexity makes it a target-rich environment—bridges consistently produce the largest single-incident losses, and the failure modes repeat with striking consistency because the core problem is architectural,” he noted.

Recurring patterns

Raz Niv, Co-Founder and CTO at onchain security platform Blockaid, told Decrypt that three technical patterns keep showing up across the year’s biggest incidents: privileged access control failures, malicious proxy upgrades where attackers swap implementation contracts for backdoored versions, and cross-chain message verification gaps.

On privileged access, Niv said the firm monitors for “anomalous ‘Role Granted’ events and unauthorized privilege escalation,” with incidents like the Echo Protocol exploit tracing back to compromised or misconfigured admin keys.

“Attackers either social engineer their way to private keys or exploit poorly designed multisig thresholds,” he added.

He pointed to failures involving privileged access controls, malicious proxy upgrades and cross-chain verification systems, saying that recent attacks are exposing deeper weaknesses in the assumptions connecting increasingly complex infrastructure.

“The common thread isn’t complexity per se,” Niv said. “It’s that each layer of abstraction (proxies, admin roles, cross-chain messaging) introduces trust assumptions that attackers methodically probe.”

AI influence

Niv said AI is increasingly transforming exploit discovery, though he cautioned that its impact is often misunderstood.

Current models are becoming increasingly effective at identifying known vulnerabilities at scale and are “automating what skilled auditors do,” he said, while warning that “the real concern isn’t AI replacing human attackers” but AI “amplifying attackers” by handling reconnaissance and freeing them to focus on more sophisticated techniques.

“The good news is defenders can use the same tools. AI-assisted monitoring and simulation is becoming essential for security teams trying to keep pace,” Niv added.

In the case of the surge in DeFi hacks, Newson pointed to a similar trend, saying “one factor that is likely a contributor, though not the sole factor, is the advances in AI.”

She added that CertiK has seen a rise in older and unverified contracts being exploited, making “the logical assumption that AI is helping find vulnerabilities.”

Similarly, Redbord said “bad actors are deploying AI at scale” across reconnaissance, social engineering, and exploit design, adding the sophistication seen in attacks like on Drift appears “consistent with AI-assisted workflows.”

TRM analysts believe North Korean operators are increasingly incorporating AI tools into their operations, with him saying, “the answer is to deploy AI on defense with the same aggression adversaries are deploying it on offense.

Above the code

Redbord said DeFi hacks are “a solvable problem,” but said that the industry needs to be more honest about where failures are actually occurring.

He noted that “audits protect against code bugs” but not against sophisticated social engineering campaigns like Drift, where North Korean proxies reportedly spent months cultivating access before the breach.

“The model that works is real-time public-private coordination,” the expert added.

Newson said 2026 may represent “an evolutionary turning point,” saying the industry is learning that cybersecurity is a “full-stack problem” spanning “AI, the DPRK, or infrastructure and personnel.”

“It doesn’t matter how perfect your math is on-chain if your human processes off-chain are vulnerable,” she said, noting the industry is increasingly shifting toward “practical, structural solutions” to address infrastructure and social-engineering risks.

Confidence hit

The damage to confidence in the DeFi space is harder to quantify but easy to observe.

The Kelp DAO exploit triggered a $6.2 billion wave of withdrawals from Aave alone, before a relief effort led by Aave CEO Stani Kulechov, dubbed “DeFi United,” raised 132,650 ETH worth roughly $303 million to backstop the bad debt.

The coordinated response shows the industry can mobilize. It also shows how much capital it takes to paper over a single bridge exploit.

Newson said the fallout depends entirely on who’s affected.

“Seasoned industry veterans may look at the last six weeks as par for the course—simply the next evolutionary norm and a harsh experience to be learned from,” she said.

She noted the impact of repeated exploits looks very different for newer market participants, warning that for users who lose significant funds, the fallout isn’t a “learning experience” but raises “existential questions” about crypto’s long-term “viability and safety,” with technical fixes often arriving too late to undo the damage.