AI-Driven Security Threats Prompt Major Shift in Crypto Protocols

John NadaBy John Nada·Apr 25, 2026·4 min read
AI-Driven Security Threats Prompt Major Shift in Crypto Protocols

The emergence of Anthropic's Mythos model is driving a fundamental shift in how crypto addresses security, focusing on infrastructure vulnerabilities.

The emergence of Anthropic's Mythos model is compelling the crypto industry to rethink its security approaches. Traditionally, decentralized finance (DeFi) has concentrated on securing smart contracts through code audits and vulnerability assessments. However, Mythos, designed to connect and identify weaknesses across multiple systems, is shifting focus toward the infrastructure that supports these protocols.

Paul Vijender, head of security at Gauntlet, emphasizes that the most significant risks now lie within the infrastructure rather than just smart contracts. Key management systems, signing services, and oracle networks are becoming focal points for vulnerabilities, as these components are often less visible and fall outside standard audit scopes. A recent security breach at Vercel, which affected multiple crypto companies, illustrates this shift. The breach was traced back to a compromised Google Workspace connection via a third-party AI tool, underscoring the potential for AI-assisted attacks on vital infrastructure.

Mythos represents a new breed of AI systems that simulate adversaries rather than merely identifying known bugs. By exploring how protocols interact, it reveals how minor weaknesses can be combined into significant exploits. This approach is gaining traction beyond crypto; financial institutions like JP Morgan are acknowledging AI-driven cyber risks as systemic and are considering tools like Mythos for stress testing. Coinbase and Binance have reportedly engaged with Anthropic to assess Mythos's capabilities.

The interconnected nature of DeFi protocols heightens the risks they face. Liquidity sharing, common oracles, and multi-layered integrations mean that vulnerabilities in one protocol can lead to cascading failures across the ecosystem. The Hyperbridge attack, where a flaw in cross-chain message verification allowed the minting of $1 billion worth of bridged tokens, exemplifies how a seemingly minor issue can escalate into a major exploit. As Vijender points out, while composability fuels innovation, it also introduces risk pathways that can spread rapidly.

Industry leaders view Mythos as an acceleration of existing adversarial dynamics rather than a radical shift. Aave Labs’ founder, Stani Kulechov, notes that DeFi has long faced well-funded adversaries, and AI models merely enhance the tools available for exploitation. The automatic execution of smart contracts and the operation of defenses without human intervention create an environment ripe for machine-speed attacks.

Although AI does not fundamentally change the landscape, it does reveal new categories of vulnerabilities. Kulechov acknowledges that AI can uncover previously deprioritized bugs, indicating that the breadth of threats is evolving. As attackers leverage AI to operate faster, the pressing question becomes whether security defenses can keep pace.

Both Gauntlet and Aave advocate for a fundamental shift in security models. Traditional audits and post-deployment monitoring, designed for human-paced threats, may no longer suffice. Vijender asserts that to counter AI-driven offenses, an AI-centric approach is necessary, prioritizing continuous adaptation and real-time simulations. This shift will require systems built on the assumption that breaches will occur.

Aave has already integrated AI into its security workflows, utilizing it for simulations and code reviews alongside human auditors. Kulechov highlights that while AI adds significant value, it complements rather than replaces human oversight. The long-term implications may not lead to disruption but divergence. As Hayden Adams, CEO of Uniswap Labs, notes, those projects that prioritize security will have a better chance of testing and hardening their systems effectively, while others risk falling behind.

The gap between secure and insecure protocols is likely to widen as AI tools enhance the capacity for stress testing and system fortification. This evolution signals a shift in the security paradigm, where the focus moves from merely eliminating vulnerabilities to continuously adapting to a landscape in which those vulnerabilities are ever-present and evolving. As the crypto ecosystem grapples with these challenges, the implications for security protocols and risk management practices will be profound, potentially reshaping the future of decentralized finance.

Scroll to load more articles