Immunefi CEO Links AI Models to Resurgence in DeFi Hacks

Cointelegraph reported that Immunefi CEO Mitchell Amador described the spread of new frontier AI models as creating a "vulnerability apocalypse" that helped drive a resurgence in DeFi hacks. Cointelegraph named the models Claude Opus 4.8 and ChatGPT 5.5 among those Amador singled out. Cointelegraph, citing DefiLlama, reported illicit actors stole more than $634 million from cryptocurrency platforms in April 2026, the article said, the highest monthly total since losses near $1.4 billion in February 2025. Cointelegraph reported Amador told the WAIB Summit in Monaco that the next three to four years will be crucial for the crypto industry unless defensive uses of these AI models and "crowdsourced security solutions" scale. Cointelegraph also reported the recent release of Anthropic's Fable 5, and that Anthropic said Fable 5 reroutes cybersecurity topics to Claude Opus 4.8.

Generative foundation models lower the technical barrier to creating exploit code, synthesizing attack chains, and automating fuzzing and vulnerability discovery. Industry observers have noted that improvements in code generation, semantic understanding of protocol logic, and rapid prompt-driven iteration reduce attacker time-to-exploit compared with traditional manual analysis. For practitioners, this implies an arms-race dynamic where defensive automation, model-aligned static analysis, and robust test harnesses must be evaluated against adversarially generated exploit sequences.

Industry coverage frames the story as part of a broader pattern where rapid AI capability growth changes attacker economics and tooling. Bug-bounty platforms, security researchers, and protocol teams are now reporting a surge in tool-assisted exploit attempts, which raises operational and triage burdens. The Cointelegraph report and DefiLlama numbers make the trend measurable at a market level, increasing pressure on security workflows, incident response, and postmortem practices across DeFi.

Watch public safety and mitigation documentation from major model providers, measurable shifts in DefiLlama monthly loss figures, and the emergence of defensive AI tooling that integrates model-based fuzzing and exploit detection. Observers should also track whether bug-bounty participation and crowdsourced disclosure programs scale in response, and whether model-release notes (for Fable 5 and others) add operational guardrails for cybersecurity-related prompts.