Ethereum cofounder Vitalik Buterin has laid out a different way to think about security: the goal is not absolute safety, but reducing the distance between what a user means to do and what the system actually does. That framing shifts the discussion for wallets, smart contracts, and software security toward intent rather than an unrealistic promise of total protection.
In a recent post on X, Buterin said “unconditional security” cannot be achieved. His point was not simply that systems are imperfect. He argued that human intention itself is complex, and that complexity cannot be captured fully in code. Even a well-built system can still misread or incompletely represent what a person wants.
A simple ETH transfer can still carry ambiguity
To explain the problem, Buterin used a basic example: sending 1 ETH to Bob. On the surface, the instruction looks straightforward. In practice, Bob may be represented by a public key, yet that key may not actually reflect the recipient the user has in mind. If a contentious hard fork is involved, even the question of which chain represents ETH can become subjective.
That is why he ties security to common sense as much as code. User intent is filtered through assumptions people make naturally, and those assumptions are hard to formalize. A system built around one narrow interpretation can miss what the user actually meant, so a single security check is not enough.
Redundancy matters more than a single line of defense
Buterin argued that strong security mechanisms need multiple ways for users to encode intent. He pointed to programming language type systems as one example. In that setting, data structures and program actions are specified together, and the code fails to compile if those elements do not line up. It is a practical way to catch mismatches early.
The same logic applies to Ethereum tools. Transaction simulations let users inspect on-chain consequences before confirming. Multisig wallets, spending limits, and post-assertions add more layers at different stages of execution. Under this model, security is treated as risk reduction through overlapping protections, not as a claim that risk can be eliminated.
LLMs can assist, but cannot define intent alone
Buterin also said AI systems such as large language models can be viewed as “shadows” of human intention. A general-purpose LLM may capture common-sense reasoning, while a model fine-tuned for one user may better reflect individual preferences and patterns.
Still, he drew a hard line on their role. LLMs should never be the sole basis for determining intent. Instead, they should sit alongside established methods and offer another perspective. In his view, combining different mechanisms increases redundancy and cuts the chance that a system drifts away from what the user truly wanted.
Security should not turn every action into friction
Buterin also pushed back on the idea that stronger security must mean more clicks and more delay. Low-risk actions should remain easy to complete. Riskier operations should face added confirmation and stricter review. The distinction matters because a secure system that is too cumbersome can distort user behavior in its own way.
His argument sketches a broader design principle for Ethereum products. Security is not a standalone feature bolted onto wallets or contracts. It is a combination of intent expression, layered verification, and usable interaction design, with the aim of keeping system behavior as close as possible to what the user meant.