Artificial intelligence is making vulnerability hunting easier and cheaper than ever, forcing blockchain ecosystems to comprehensively overhaul their cybersecurity strategies. In an in-depth analysis on Bankless, author William M. Peaster observes that the crypto world is splitting into two opposing schools of thought to counter AI-equipped attackers: upgradable agility and code minimalism.
This pressure follows a series of major incidents, from Bitcoin Core changing its security policy due to AI to Initia halting its blockchain following a suspected $2.6 million exploit. Data from DefiLlama recorded 112 attacks causing losses exceeding $1 million in 2026, far surpassing the 70 incidents recorded in all of 2025. Anthropic’s research also points out that AI models’ smart contract exploitation capabilities double every 1.3 months.
The agile school: Ready to intervene and rapidly upgrade
The first response is to maintain the ability to patch bugs and intervene promptly in the network. The most prominent example is the Starknet community’s decision to vote in favor of exploring a plan to become an independent, quantum-resistant L1 chain. The development team argued that they could not afford to wait for Ethereum’s lengthy decision-making process to upgrade cryptography to safeguard user assets.
However, agility comes at a steep price. Admin keys and contract upgrade mechanisms are themselves lucrative targets for hackers. Moreover, constantly changing operational rules can inadvertently introduce new logic bugs, similar to how Ethereum recently had to postpone the Glamsterdam testnet due to gas fee mechanism conflicts.
Minimalism and the rise of formal verification
On the other hand, minimalist developers choose to minimize the attack surface as much as possible. The key benchmark here is the “walkaway test”: can a protocol continue to operate reliably and securely even if its founding team completely vanishes tomorrow? When code is concise, highly deterministic, and immutable, hackers and AI are left with very little room to maneuver.
This is also why Ethereum’s cryptographic roadmap is shifting toward a hash-only structure, as hashes lack the internal mathematical structure that machine learning algorithms can exploit. A breakthrough emerged when Anthropic used Claude for formal software verification to automate mathematically proving that code cannot crash on any arbitrary input.
Historically, formal verification was prohibitively expensive and demanded decades of expert labor. As AI drastically lowers this mathematical overhead, small, single-purpose smart contracts stand to benefit the most. In a future where proving absolute safety becomes feasible, immutability will no longer be seen as rigid vulnerability, but rather the most formidable shield against automated exploit hunters.