The crypto attacks struck just over two weeks apart in April, allowing hackers to steal nearly $600 million combined. The breaches sparked a wave of investor withdrawals from one major platform and forced another company into collapse.
What has most concerned cybersecurity analysts, however, is the method behind the attacks. Investigators believe the hackers — suspected to be groups tied to North Korea — used artificial intelligence to identify targets and craft sophisticated exploits, according to blockchain intelligence firm TRM Labs.
Experts say the operations showed an unprecedented level of sophistication, suggesting AI played a major role in helping the attackers execute the massive crypto heists. TRM investigator Nick Carlsen said the tactics represented a significant evolution in cybercrime capabilities.
Artificial intelligence falling into the hands of cybercriminals could dramatically increase the risks facing the crypto industry, which has already suffered billions of dollars in losses from hacks in recent years. Because cryptocurrency platforms rely heavily on blockchain infrastructure, the sector remains especially vulnerable to digital attacks.
The danger became clear after one of the April breaches, when investors withdrew nearly $9 billion within just two days from a lending protocol allegedly used to move stolen funds. The massive outflow highlighted how rapidly trust can collapse in the crypto market — even when the platform itself was not directly attacked.
“There is virtually no margin for error when it comes to security today,” said Nicholas Smart of blockchain investigations firm Crystal Intelligence.
Adding to concerns is Mythos, an advanced AI model from Anthropic that has reportedly been restricted from broad release because of cybersecurity fears. While there is no proof the recent hackers used the system, researchers warn it may only be a matter of time before criminals gain access to increasingly powerful AI technology. Anthropic’s own studies suggest even current AI agents are already capable of carrying out sophisticated exploits.
Security experts also fear AI could make high-level cybercrime accessible to far more people. “Before AI, there may have been only a small group of elite hackers,” said Niv Yehezkel of Chainalysis. “Now, almost anyone with a subscription can operate at the level of an elite hacker.”
Decentralized finance, or DeFi — a $130 billion segment of the crypto industry where users trade, borrow, and lend digital assets through automated protocols — has become one of the sector’s biggest security weak points. In April, DeFi-related exploits surged to a record high, nearly doubling compared to the previous month, prompting projects across the industry to urgently reinforce their cybersecurity defenses.
Although many of the attacks involved relatively small amounts, security specialists say the sharp increase highlights how cybercriminals are becoming more efficient at detecting software vulnerabilities and rapidly creating exploits. Experts believe widely accessible AI tools are likely playing a major role in accelerating those attacks.
Cybersecurity researchers caution that proving hackers used AI is difficult. Instead, investigators typically assess the sophistication of the operation, the tactics involved, and how challenging it would have been to identify and exploit the target manually. According to several experts interviewed, the sudden spike in crypto heists is itself a strong sign that attackers are increasingly relying on artificial intelligence.
“With AI, the cost of vulnerability discovery is rapidly approaching zero,” said Aneirin Flynn, CEO of Security Audit Firm, Failsafe. He noted that what once took hackers months to uncover in blockchain protocols can now be done in days — or even hours — with the help of artificial intelligence.
The cybersecurity risk posed by AI extends well beyond crypto. In November, Anthropic reported that attackers had manipulated its Claude model in attempts to breach around 30 organizations, including major tech firms, financial institutions, and government agencies, succeeding in a limited number of cases. The company did not disclose the identities of the targets.
When executives at Anthropic realized that its Mythos model was significantly more capable of executing cyberattacks than earlier systems, they opted to restrict its early rollout to a small group of major global tech firms. The goal was to allow those companies to test the model against their own systems in a controlled environment. Several large banking institutions have also begun evaluating Mythos in similar security-focused trials.
Patchwork of Code
Decentralized finance (DeFi), however, remains particularly exposed. Unlike traditional banking systems, regulatory and security oversight in the sector is fragmented. In conventional finance, regulators routinely conduct stress tests on major banks to evaluate their cyber resilience, while institutions can block or reverse suspicious transactions. In contrast, blockchain transactions are irreversible, and attackers often have multiple pathways to launder and move stolen assets beyond recovery.
DeFi has grown in popularity among crypto investors seeking yield and operates as a network of interconnected blockchain protocols powered by self-executing smart contracts. These systems allow users to transfer and deploy digital assets without relying on centralized intermediaries. However, investment in security infrastructure varies widely across projects, leaving many platforms unevenly protected against increasingly sophisticated threats.
That leaves attackers with a broad range of potential targets across the ecosystem, while also increasing the risk that the fallout from a single breach spreads and exposes other platforms indirectly connected to the attack.
The two major April incidents highlighted both dynamics. The first targeted the derivatives exchange Drift Protocol, draining more than $280 million. In a post-incident report released days later, Drift said the attackers had spent months cultivating trust with contributors while posing as a quantitative trading firm. They ultimately deceived staff into approving malicious transactions.
The operation was also notable for its complexity. The attackers created a fake token and engineered an artificial trading history to make it appear legitimate, tricking Drift’s systems into accepting it as valid collateral.
Drift Protocol was forced to suspend operations and is now planning a relaunch after receiving an injection of stablecoins from Tether. Meanwhile, another DeFi project, Carrot, which had exposure to Drift’s ecosystem, announced on April 30 that it would shut down as a direct consequence of the incident.
The second major attack hit Kelp DAO, where hackers targeted a cross-chain “bridge” — a protocol designed to connect different blockchains. While parts of the exploit, which netted nearly $300 million, remain unclear, the aftermath proved even more disruptive due to how the stolen funds were laundered.
In an unconventional move, attackers used a large portion of the stolen assets as collateral to borrow funds on Aave, the largest DeFi lending platform. This triggered concerns that Aave was holding compromised collateral, prompting a wave of withdrawals that spilled over into unrelated platforms. The situation escalated to the point where Aave ultimately required intervention to stabilize.
The Drift Protocol and Kelp DAO breaches differed in their execution. The Drift attack appeared to lean heavily on social engineering, with attackers deceiving individuals into granting access to restricted systems. But experts say the overall sophistication went well beyond previous cyberattacks, prompting speculation that artificial intelligence may have played a role in areas such as planning and exploit design.
“I highly suspect that North Koreans used AI to engineer both” incidents, said Nick Carlsen, a former FBI analyst now working at TRM Labs. “This is all stuff North Korea never used to do.”
There have also been efforts to evaluate how capable AI agents are at identifying blockchain vulnerabilities and executing attacks. In December, Anthropic published research suggesting that more than half of blockchain exploits in 2025 — “presumably carried out by skilled human attackers” — could potentially have been executed autonomously by AI systems.
Research from Anthropic found that what it described as “potential exploit revenue” was doubling roughly every 1.3 months, while the cost of executing successful hacks was falling sharply. The researchers concluded that “profitable autonomous exploitation can happen today,” warning that AI-driven attack capabilities were already economically viable.
Anthropic declined to say whether it had repeated the experiment using its Mythos model.
A separate assessment by two engineers at a16z produced more mixed results. According to a April 28 blog post by Daejun Park and Matt Gleason, an AI system trained on previous DeFi exploits was consistently able to identify vulnerabilities in protocols, but struggled to fully construct a profitable end-to-end attack. The researchers noted the work was conducted before the release of Mythos and said they intended to test the model once access becomes available.
Building Defenses
Within the crypto industry, urgency around cybersecurity is intensifying. Aneirin Flynn of Failsafe said some clients are now deploying monitoring systems that continuously scan devices connected to their networks — including laptops and mobile phones — to detect unusual activity and alert administrators in real time.
Yuan Han Li of Blockchain Capital has also advocated for stronger “circuit breaker” mechanisms that can pause or limit transactions once certain thresholds are exceeded, effectively creating time to respond to potential exploits. One trading platform, Jupiter, already uses a similar system and is expanding its deployment, according to co-founder Siong Ong.
Aave is also broadening its risk framework to incorporate cybersecurity considerations into how collateral is assessed, said chief legal and policy officer Linda Jeng.
Despite these defensive efforts, some experts remain skeptical that purely defensive strategies will be enough against increasingly AI-assisted attackers. Nick Carlsen of TRM Labs argued that conventional security approaches may not suffice against adversaries like North Korea-linked groups using advanced tools. In his view, the only viable response may be to adopt more aggressive countermeasures against stolen funds. “You don’t win this kind of campaign playing defense,” he said. “They need to be hacked.”
Source: Business-standard.com Edited By Bernie
