Whalesbook
Infrastructure Becomes Top Security Concern
Cryptocurrency security is changing significantly. Anthropic's new AI model, Mythos, is prompting the industry to look beyond traditional smart contract defenses and confront vulnerabilities in critical infrastructure. This marks a major shift from years spent auditing code and cataloging common exploits.
Prioritizing Infrastructure Over Code
"The bigger risks sit in infrastructure," stated Paul Vijender, head of security at risk management firm Gauntlet. He notes that AI-assisted attacks targeting human elements and infrastructure layers are now a greater risk than smart contract exploits. These vital components include key management systems, signing services, oracle networks, and cryptographic layers often outside the scope of standard audits. An example is a recent incident at Vercel, a web infrastructure provider used by crypto firms, where compromised credentials risked customer API key exposure.
How AI Models Like Mythos Find Weaknesses
Mythos represents a new type of AI designed to simulate adversaries by finding and connecting weaknesses across complex systems. Rather than just scanning for known bugs, these models test how small vulnerabilities can be combined into real-world exploits. This capability has drawn interest outside crypto, with institutions like JP Morgan increasingly viewing AI-driven cyber risk as systemic. Both Coinbase and Binance have started testing Mythos, with early findings pointing to weaknesses in behind-the-scenes systems that protect keys and manage inter-system communications.
Interconnected Systems Amplify Risk
In a system built on composability, where decentralized finance (DeFi) protocols connect, infrastructure vulnerabilities can spread quickly. The same interconnectedness that drives DeFi's capital efficiency also creates pathways for risk to spread. AI can now map these dependencies at scale, turning isolated exploits into widespread failures that cascade across the ecosystem.
New Defenses Needed Against AI Threats
Industry leaders recognize this evolution. While some see AI as an acceleration of existing adversarial dynamics, others see it as a necessary advancement. The traditional model of pre-deployment audits and post-deployment monitoring is challenged by the speed of AI-driven threats. "To defend against offensive AI, we will need to take an AI-centric approach where speed and continuous adaptation are essential," Vijender noted. This includes continuous auditing and real-time simulation. Protocols that prioritize security and leverage AI for stress-testing are expected to widen the gap between secure and insecure projects, fundamentally changing the security landscape.