Ronghui Gu, co-founder and CEO of blockchain security firm CertiK, said this week that DeFi attackers are using AI to outspend and outmaneuver the teams trying to stop them. In remarks attributed to an interview with The Block on May 14 (the full article was behind a paywall and could not be independently verified at time of publication), Gu argued the core problem is no longer poorly written smart contract code. Instead, attackers have pivoted to operational security gaps and supply chain weaknesses, targeting the humans and infrastructure around a protocol rather than the protocol itself. CertiK, valued at $2 billion and backed by Sequoia, Coatue, and Goldman Sachs, audits code for more than 5,000 enterprise clients and claims to have secured over $600 billion in digital assets.

The numbers behind Gu's warning are stark. DeFi lost more than $770 million across 47 separate incidents in the first four months of 2026 alone, a 68 percent increase in attack frequency compared to the same period in 2025, according to CryptoTimes. April 2026 recorded the highest number of crypto hacking incidents in any single month in history, with between 28 and 30 confirmed exploits. More than 40 DeFi protocols have shut down so far this year. The carnage triggered a mass withdrawal from the sector: over $13 billion in total value locked (TVL, the total assets deposited in DeFi protocols) fled the space within 48 hours following the April attack wave, including $8.4 billion from lending platform Aave alone. That pace represents a sharp acceleration from an already worsening trend: 2025 saw $3.35 billion lost across 630 incidents for the full year, itself a 37 percent increase over 2024's $2.45 billion.

The shift in attack methodology is documented across multiple security research sources. CertiK's 2025 Hack3d report found that phishing was the single largest attack vector last year, accounting for 248 incidents and $722 million in losses. Private key compromises, where attackers gain control of the cryptographic credentials that authorize transactions, accounted for 88 percent of all stolen funds in the first quarter of 2025. Off-chain attacks overall, meaning those targeting people and infrastructure rather than blockchain code, were responsible for 80.5 percent of stolen crypto funds in 2024, according to security research firm DeepStrike. Yet Halborn's analysis of the top 100 DeFi hacks found that only 19 percent of victimized protocols used multi-signature wallets (which require multiple parties to approve transactions) and just 2.4 percent used cold storage (offline key management). The February 2025 Bybit breach illustrated exactly how these conditions combine into catastrophe. Lazarus Group, the North Korean state hacking collective, socially engineered a developer at Safe{Wallet}, a third-party transaction signing service used by Bybit. The attackers stole AWS session tokens (temporary, time-limited access grants), bypassed multi-factor authentication, and injected malicious code into the Safe{Wallet} interface. Bybit's own staff then unknowingly signed off on a transaction that moved roughly 401,000 ETH to attacker-controlled wallets. The total loss reached approximately $1.5 billion, still the largest single crypto theft in history. North Korea-linked groups have since come to dominate the threat landscape: TRM Labs attributes 76 percent of all 2026 crypto hack losses through April to DPRK-affiliated actors.

CertiK's response on the defensive side is a tool called AI Auditor, launched in April 2026. The firm says it achieved an 88.6 percent detection rate against 35 real-world Web3 security incidents from 2026 using a multi-stage parallel scanning architecture, and that it has flagged more than 180,000 vulnerabilities in blockchain code to date. Gu framed the tool's purpose carefully: "The question is no longer simply whether AI can find vulnerabilities, but whether it can genuinely help development teams surface the security issues worth addressing, earlier." Hudson, CertiK's head of ecosystem, put it more plainly: "People aren't securing their code enough."

The implications land hardest in regions where crypto adoption is growing fastest but security infrastructure is thinnest. India ranked first globally in the Chainalysis 2025 Crypto Adoption Index, while Pakistan moved to formalize its crypto sector through a newly established Pakistan Crypto Council and a planned dedicated regulatory authority, the Pakistan Virtual Assets Regulatory Authority (PVARA). In Sub-Saharan Africa, on-chain transaction volume exceeded $205 billion between July 2024 and June 2025, a 52 percent year-over-year increase, and four African countries now appear in the top 20 of the 2026 Global Crypto Adoption Index. Nigeria holds a near-top position in that ranking, while Ethiopia and Kenya made their debuts in the top 20 this year. Nigeria alone saw roughly $250 million extracted from retail users through CBEX, a fraudulent investment scheme that used crypto infrastructure and relied on social engineering rather than code exploits. In markets where developer teams run on tight budgets, users interact primarily through mobile interfaces, and regulatory frameworks for recovery are still being built, AI-generated phishing in local languages and deepfakes of known local figures represent exactly the low-cost, high-return attack surface Gu is describing.

Whether defensive AI can close the gap remains an open question. CertiK's own figures show that even a high detection rate at the code level does not address the human and operational vulnerabilities now driving most losses. The single largest individual hack of 2026 was the Kelp DAO breach on April 18, a $293 million loss triggered by a forged LayerZero bridge message rather than any flaw in Kelp DAO's own code. The Drift Protocol breach, a $285 million incident that unfolded over six months of social engineering targeting admin key holders rather than any code flaw, reinforces the same lesson. The hardest attack surface to audit is people.