Artificial Intelligence Forces a Global Security Reckoning Across Crypto and Cybersecurity

Artificial intelligence is growing at an exponential pace in its ability to discover software vulnerabilities, and the broader cryptocurrency industry—alongside the wider technology sector—is rapidly reorganizing around this sobering new reality. The foundational layer of the digital asset economy caught an immediate glimpse of this shift when Bitcoin Core officially merged a critical security policy change designed specifically to address bugs "that can be found with a simple prompt to an LLM."

Rather than allowing these potential security vectors to sit quietly under traditional multi-month embargoes that could last up to a full year, Bitcoin developers will now rush to fix and ship these patches as rapidly as possible. As Antoine Poinsot, the author of the policy change, noted regarding the update, large language models are drastically reducing the barrier to entry required to find certain classes of deeply hidden code issues. This single adjustment highlights a broader awakening across software engineering: security through obscurity is no longer a viable defense when automated, tireless AI models can analyze source code repositories in seconds.

The broader digital landscape is already feeling the fallout of this technological leap. Recently, two critical PHP vulnerabilities discovered with the assistance of advanced artificial intelligence models left approximately 1.3 million internet-exposed hosts vulnerable to potential exploitation. Shortly after, the Initia network was forced to halt its entire operation after a malicious actor exploited what cofounder Zon described as a "Day 0 bug" to mint unauthorized tokens and bridge out approximately $2.7 million. While definitive confirmation of AI utilization in the Initia exploit remains unverified, industry observers note that it is increasingly safe to assume artificial intelligence played a role in some capacity, given the sophistication and speed of modern threat actors.

Data compiled by DefiLlama’s hacks database illustrates the severity of this evolving threat landscape. The database currently logs 112 hacks exceeding $1 million in losses so far in 2026, marking a dramatic escalation compared to the 70 major exploits recorded during the entirety of 2025. Furthermore, recent smart contract research published by Anthropic revealed that frontier AI models have seen their simulated exploit revenue double roughly every 1.3 months throughout 2025.

Faced with an ecosystem where attackers possess tireless, constantly improving automated bug hunters on tap, the cryptographic community is actively converging on two fundamentally very different strategic answers to ensure long-term survival.

The agility approach

The first major philosophy gaining traction across the industry is the agility approach, which emphasizes maintaining the operational nimbleness required to patch, upgrade, or halt networks swiftly when threats emerge.

Starknet stands out as one of the clearest examples of this philosophy in action. In recent community governance votes, stakeholders agreed to research the possibility of transforming into an independent, post-quantum Layer 1 network, driven partly by the desire to upgrade its underlying cryptography on its own independent schedule rather than waiting for broader ecosystem consensus. As StarkWare articulated in public statements, developers cannot afford to depend on Ethereum moving fast enough to ensure their immediate safety. StarkWare CEO Eli Ben-Sasson now pegs the practical arrival of quantum computing threats at roughly two years away, arguing that rapid advancements in artificial intelligence are drastically accelerating the timeline.

Ethereum researchers are also on high alert regarding these developments. Earlier in the month, core researcher Justin Drake publicly urged the broader crypto industry to begin planning for "bunker mode," a defensive posture involving the movement of funds to fresh, structurally safer addresses following a wave of AI-driven mathematical breakthroughs.

In practice, however, maintaining extreme agility carries its own distinct set of risks and costs. Every administrative key, multisig arrangement, and modular upgrade path introduces another surface area for potential attacks or dependencies. Moreover, altering operational rules under pressure can inadvertently introduce novel bugs of its own. This risk was underscored when Ethereum developers decided to postpone the Glamsterdam upgrade on the Hoodi testnet after research indicated that proposed gas changes could fundamentally alter how already-deployed smart contracts behave in production.

The minimalism approach

The alternative philosophical answer to the AI threat is to drastically shrink the attack surface altogether through radical minimalism. This concept was encapsulated by Ethereum developer Ross, known online as z0r0z, who argued that developers must fundamentally pivot toward simpler, highly constrained systems.

This mindset is a direct expression of the walkaway test philosophy in software engineering, which questions whether a system can continue operating safely and autonomously even if its original creators disappear tomorrow. Small, deterministic pieces of code that can be easily and exhaustively verified by human eyes or automated tools leave fewer places for architecture to break and fewer third-party entities to trust. As Ross emphasized in subsequent discussions, developers must refocus their efforts entirely on writing simpler codebases.

The exact same defensive instinct is increasingly visible in core Ethereum cryptography. Ethereum co-founder Vitalik Buterin noted that the network’s long-term cryptographic roadmap has transitioned toward a "hash-only" design philosophy, in part because cryptographic hashes are specifically engineered to possess no exploitable mathematical structure that could be systematically cracked by AI-accelerated mathematics.

Nevertheless, minimalism has its own inherent trade-offs. Public, immutable codebases are equally readable by AI-driven attackers, and once such code is deployed live to a decentralized network, it cannot be easily patched. Consequently, the minimalism approach works most effectively when the underlying code is small and simple enough to be checked exhaustively before deployment.

Where AI proofs come in

This limitation brings formal verification—the rigorous process of mathematically proving that a piece of code cannot violate specific security rules—to the forefront of modern software security.

Historically, formal verification has been a brutally expensive and labor-intensive endeavor. Building the seL4 operating system kernel required two person-years of engineering effort, but took an astonishing 20 years to formally verify. However, in groundbreaking research published alongside Theorem, Anthropic utilized the Claude AI model to prove that 239 distinct entry points across 123 open-source libraries cannot crash on any input up to 64 megabytes, operating under specific structural assumptions. Remarkably, the AI accomplished this monumental verification task in just a few weeks.

Despite these promising breakthroughs, significant hurdles remain. Mathematical proofs scale directly with the complexity of the underlying code. A typical proof generated during Anthropic’s research effort spanned approximately 600,000 lines of proof code for a program consisting of roughly 18,000 machine instructions, consuming hundreds of millions of AI tokens in the process. Furthermore, a formal proof only covers the specific properties and invariants explicitly defined by the developer, rather than guaranteeing immunity to every conceivable bug.

Yet, a closer examination reveals which sector of the industry stands to benefit most from drastically cheaper and faster formal verification: small, single-purpose software libraries and smart contracts. This is precisely the type of code that proponents of minimalism aim to deploy onchain. If artificial intelligence continues to drive down the cost and friction associated with formal verification, the concept of unpatchable code will matter significantly less, because the code will be mathematically guaranteed not to need patching in the first place.

Agility will undeniably remain a vital necessity for massive, evolving systems such as Layer 1 networks and modular rollups that require continuous adaptation. Even so, within this rapidly evolving artificial intelligence paradigm, the walkaway test is increasingly resembling a definitive security north star rather than a theoretical decentralization ideal. As software engineering enters this uncharted territory, the entire digital asset ecosystem will need to adapt its security models accordingly.