Smart contracts

Ethereum's developer documentation, updated 25 February 2026, describes a smart contract as an account containing code and state. It executes only when called. Deterministic execution means every executing node reaches the same result from the same inputs. A contract cannot fetch off-chain facts itself. Those arrive through signed transactions or an oracle. Contracts are not deletable by default, and interactions are irreversible. NIST IR 8202, published in October 2018, also makes clear that these programs need not perform financial functions.

The smart contract label does not make a program a legal agreement, an independent actor or a guarantee of correct behavior. A smart contract can follow flawed instructions exactly. Its logic may be replaceable, and visibility depends on the platform. Daml on Canton restricts what each party sees. Many contracts publish no verified source code. Ethereum's nontechnical documentation, updated 1 July 2026, calls blind signing a major risk: a user approves an interaction without understanding its effect.

Compiling Solidity source or Vyper produces EVM bytecode and a JSON ABI describing functions. Calls start with a 4-byte function selector. A deployment transaction sends initcode without a recipient, not to the zero address. Initcode returns runtime code. CREATE uses creator and nonce, while CREATE2 uses creator, salt and initcode. Ethereum mainnet limits on 2 November 2026 are 24,576 runtime bytes and 49,152 initcode bytes. In the 2 November 2026 lab, storing code accounts for 81.6% of deployment gas.

Each transaction incurs gas costs, even when a revert undoes state changes. An event writes a log that contracts cannot read. STATICCALL invokes a view function without allowing state changes. DELEGATECALL runs another contract's code against the caller's storage. Since Dencun on 13 March 2024, SELFDESTRUCT deletes code only within its creating transaction under EIP-6780. ERC-4337 uses bundlers and EntryPoint, with optional paymasters covering fees. EIP-7702 has let accounts with keys set code since Pectra on 7 May 2025. The EVM cannot access networks or files.

EIP-7954 raises runtime code to 65,536 bytes and initcode to 131,072 bytes, or 64 KiB and 128 KiB. This Ethereum improvement proposal is scheduled for Glamsterdam, live on Sepolia since 6 October 2026. On 22 October 2026, the Ethereum Foundation said core developers had postponed Glamsterdam on the Hoodi testnet to investigate EIP-8037 interactions with mainnet smart contracts. There is no mainnet date on 2 November 2026. Solidity's unreleased 0.8.39 already warns about these limits when targeting the Amsterdam EVM version.

Ethereum documentation calls deployed code immutable by design, but distinguishes upgradeability from mutability. Migration, data separation, strategy contracts and proxies change how later behavior is reached. ERC-1967 reserves slots for implementation, admin and beacon addresses. UUPS puts upgrade logic in the implementation. A beacon changes several proxies in one transaction. ERC-2535 diamonds distribute functions across facets, while ERC-7201 defines namespaced storage. In OpenZeppelin Contracts 5.7.0, checked on 2 November 2026, the stated recommendation shifts toward UUPS. None of these mechanisms edits the original deployed code.

Users must trust whoever controls future logic. A multi-signature wallet or DAO governance process can distribute upgrade authority. A timelock gives users time to exit but delays urgent fixes. Initialization needs checks, and existing variable order and types must stay unchanged. The 2 November 2026 lab accepts a reordered layout, then reads an owner's address as a number. An uninitialized implementation also admits a stranger as owner. That takeover does not affect the UUPS proxy in this run, because its context check rejects the attempted upgrade.

EVM contracts keep code and state at one address. Solana programs, including those written with Anchor, keep mutable state in separate accounts. Move's typed resources constrain how assets are handled. Sui organizes storage around objects with unique IDs. Cardano and Bitcoin check spending conditions. Canton runs Daml smart contracts, exposing only each party's part of a transaction. Clarity disallows reentrancy at language level. Polkadot's Rust contract language project has not been maintained since January 2026. Token interfaces also differ, including ERC-20 token contracts on Ethereum.

Usage figures measure different things. Etherscan counts 939,820 verified contracts on 2 November 2026. Blockscout records 2,245,546 new contracts in October 2026, without clarifying whether creation by other contracts is included. DefiLlama showed $85.094 billion in total value locked on the evening of 2 November 2026, and $44.41 billion on Ethereum in its chain data. TVL measures deposits in tracked DeFi protocols, not every asset held by a smart contract. These counts therefore describe different slices of activity, rather than a shared measure of adoption.

The OWASP Smart Contract Top 10 for 2025 covers access control, price oracle manipulation, logic errors, reentrancy and other failure classes. The 2026 edition, current on 2 November 2026, uses 2025 incidents and survey data. It adds arithmetic errors and proxy and upgradeability vulnerabilities, while dropping insecure randomness and denial of service. The SWC registry is no longer actively maintained, with no new entries since 2020. Its content entered the EEA EthTrust Security Levels specification.

The DAO's recursive withdrawal drained over 3.6 million ETH in June 2016. Ethereum's fork on 20 July 2016 moved funds into a withdrawal contract. Parity's 19 July 2017 incident exposed unchecked initialization through DELEGATECALL. Its 6 November 2017 library deletion froze dependent wallets. Under today's EIP-6780 rules, the same kill would leave the library's code in place. Curve pools lost over $50 million on 30 July 2023 (LlamaRisk). A Vyper compiler bug defeated reentrancy protection, so correct source code cannot rule out compiler faults.

Sources use different methods and revise earlier years. Chainalysis attributes 43.8% of 2024 theft value to private key compromises. From CertiK's original figures, phishing and compromised keys account for 80.6% of 2024 losses. Supply chain attacks and phishing account for 64.8% in 2025. These calculations attribute most losses to causes other than smart contract code bugs. Chainalysis' December 2025 report includes Bybit's $1.5 billion theft. CertiK's 2025 report uses a separate supply chain category, so labels should stay attached to sources.