TLDR
Web3 exploit losses have fallen sharply in the last couple of years, helped by stronger security practices and emerging AI tools, even as AI also empowers attackers.
- DeFi exploit losses dropped about 74% from their 2022 peak, and May 2026 losses were under $100 million after an AI-fueled April spike.
- Advanced AI models now find deep protocol bugs and infrastructure flaws, pushing projects toward continuous, AI-augmented audits and more rigorous code verification.
- Attackers are also weaponizing AI, so the landscape is shifting into a full security arms race rather than a simple AI makes crypto safe story.
Deep Dive
1. Exploit Losses Have Plunged
Immunefi reports that DeFi protocol exploit losses fell from about $2.62 billion in 2022 to $680.3 million in 2025, a 74% drop, with median loss per exploit down 75% to $1.5 million by 2025. This coincides with fewer catastrophic bridge and reentrancy exploits and better access-control standards across major chains and protocols, according to its 2026 ecosystem audit.
More recently, CertiK estimates crypto exploit losses in May 2026 at roughly $68.3 million, compared with around $650 million in April, when a handful of very large DeFi attacks drove the totals. May was the third month in 2026 with losses below $100 million, suggesting structural improvement rather than a one-off fluke.
Losses are still large in absolute terms, but the worst-case DeFi blowups are becoming less frequent as defenses mature.
2. AI As A New Defensive Layer
AI is increasingly central to those defenses. Anthropics Claude Opus 4.8 was used to uncover a four-year-old bug in Zcashs Orchard pool that could have allowed unlimited counterfeit ZEC creation, leading to an emergency patch and protocol upgrades within days.
Frontier models like Claude Mythos and similar systems are now being used to scan browsers, operating systems and blockchain code, identifying tens of thousands of vulnerabilities before they are exploited. Security firms and protocol teams are moving toward continuous, AI-augmented review and, longer term, formal verification, where code is mathematically proven to satisfy certain properties.
When you evaluate a protocol, its use of ongoing audits, automated scanners and formal verification is becoming as important as whether it has a single legacy audit report.
3. AI Attack Surface And Risks
The same AI capabilities are boosting attackers. Anthropic found that about 67% of banned accounts over a recent year used AI to prepare cyberattacks, including writing malware and automating post-compromise techniques. Analysts also link an April 2026 spike of roughly $630 million in crypto hacks to AI-assisted exploit development.
Researchers warn of AI-generated zero-day exploits against common software, and maintainers report being flooded with low-quality, AI-written bug reports that can bury the real issues. Meanwhile, CertiK highlights that attackers are expanding beyond on-chain exploits to AI-enhanced phishing and even physical wrench attacks that target keyholders directly.
Falling headline losses do not equal safety; risk is shifting toward more sophisticated, AI-driven attacks and toward weak links like bridges, private keys and shared infrastructure.
Conclusion
AI is reshaping Web3 security by making it easier to discover and fix serious vulnerabilities, which contributes to the clear decline in aggregate exploit losses. At the same time, AI is lowering the barrier for sophisticated attacks, turning crypto security into an arms race between automated defenses and automated offense. For users and builders, the edge now lies in favoring ecosystems that invest in continuous, AI-augmented security and robust key and bridge practices, rather than assuming lower recent loss numbers guarantee safety.
CoinsKid.Com