Why a blockchain needs an oracle to price anything
A blockchain is a sealed room. It agrees on its own history, and it agrees on nothing else. Ask it for the dollar price of ether, the result of an election, or whether a parcel was delivered, and it has no answer, because none of that lives inside the ledger. Every validator has to reach the exact same result when it runs a transaction. If two nodes each fetched a price from the open internet a second apart, they might get slightly different numbers, and the network would fall out of agreement.
So the chain trusts only what is already recorded on it. That rule keeps it honest. It also makes it blind. A lending app cannot liquidate a loan without knowing what the collateral is worth, and a prediction market cannot pay a winner without knowing who won. The bridge between the sealed room and the outside world is the oracle.
What an oracle actually does
An oracle is a service that takes information from off the chain, packages it, and writes it on-chain in a form a contract can read. Picture a courier with a signature. It does not decide what happens next. It delivers the number and vouches for where it came from.
Two delivery styles dominate. In the push model, the oracle watches a source and posts an update whenever the value moves past a set threshold, so the latest figure is always sitting on-chain and ready. In the pull model, the data is signed and kept off-chain, and an application grabs it and posts it only at the moment it needs it, which is cheaper and faster for markets that move by the second. Both feed the same kind of consumer, the smart contracts that run on their own once the data has landed.
Prices are the loudest use, not the only one. Oracles also carry sports scores to betting markets, weather data to crop insurance, proof that a real-world asset exists to the token standing in for it, and tamper-resistant randomness for on-chain games and fair token draws. The shape of the request changes. The job does not, which is to bring a fact the chain cannot check into a place the chain can act on.
The single point that keeps breaking
The first oracles were one server run by one team. That is fast and simple. It is also a gift to an attacker. Corrupt that one feed, or wait for it to go down, and every contract downstream inherits the lie. A wrong price does not stay a display error in these systems. It becomes a real trade, a real liquidation, a real withdrawal that nobody can claw back.
The answer was to spread the job out. A decentralized oracle network pulls the same value from many independent operators, discards the outliers, and publishes the median. To force a false number through, an attacker now has to capture a majority of those nodes at once rather than bribe a single machine. Running it costs more and moves slower, and that cost buys something worth paying for, which is the absence of one weak link deciding what millions of dollars are worth. The design goal is dull on purpose. No single operator, and no single outage, should be able to move the number the whole market reads.
How a bad price turns into free money
The most common oracle attack does not break any cryptography. It feeds an honest oracle a real price from a market the attacker briefly owns.
The tool is the flash loan, which lets anyone borrow an enormous sum with no collateral as long as it is paid back inside the same transaction. The attacker borrows, then buys or dumps a thin trading pool to shove its price far from reality, and any contract reading that pool as its price source treats the fake number as gospel for one block. In October 2022 a trader named Avraham Eisenberg did exactly this to Mango Markets. He started with about $10 million, inflated his own position past $400 million on paper, and borrowed against that phantom value until roughly $117 million had drained from the platform. He was not the only one. Security firm Cyfrin counts more than $403 million taken across over forty price manipulation attacks in 2022 alone.
The problem has not aged out. By some tallies price manipulation now drives one in eight crypto hacks, which is why the choice of price source is treated as a security decision and not a plumbing detail. Builders lean on two defenses. A time-weighted average smooths the value over many blocks, so a one-block spike barely registers, though it lags when the market genuinely moves. Or a contract pulls from a decentralized feed whose median already reflects deep markets rather than one shallow pool. Even careful teams cut it fine, as when a dormant Notional Finance contract leaked $1.7 million after sitting untouched. The pattern repeats. The price source is the attack surface.
Who runs the feeds now
One name sits on top of this market. As of late August 2026, DefiLlama credited Chainlink with securing about $39.4 billion across 526 protocols, roughly 63 percent of all the value the tracked oracles stood behind. The runner-up, Chronicle, secured about $7.5 billion. Chainlink's own dashboard counts far more, near $110 billion, because it includes activity DefiLlama does not index.
That concentration is easy to read as fragility, and to a point it is. Yet the fees underneath are thin for what they guard. Chainlink took in roughly $63 million over a trailing year against tens of billions in secured value, a fee-to-value ratio near 0.16 percent. Rivals price the work differently. Pyth leans on the pull model and sources quotes straight from trading firms and exchanges, which suits fast derivatives venues. A copper perpetual on Kalshi that settled against a crypto oracle rather than the traditional metals market is the kind of product this newer plumbing makes possible. Chainlink has also pushed past pure price feeds into moving value between chains, reporting about $4.9 billion of cross-chain volume in the second quarter of 2026, a figure it says grew more than threefold from a year earlier. The feeds have become infrastructure, and most users never learn their names.
Frequently asked questions
Is an oracle part of the blockchain?
No. An oracle sits beside the chain. The contract lives on-chain, and the oracle hands outside data to it, but the two are separate services with separate trust assumptions.
Why not let one trusted company post the price?
Speed and simplicity, and that is the trap. A single provider is one outage, one subpoena, or one bribe away from feeding every downstream contract a bad value. Sharing the work across many operators is slower, yet it removes the one weak link.
Can an oracle be wrong when nobody is cheating?
Yes. If the underlying market is thin, or a venue the oracle samples suffers its own glitch, the honest median can still drift from fair value. Careful designs sample many exchanges and weight them by volume to blunt that risk.
What separates a push oracle from a pull oracle?
A push oracle keeps a fresh value sitting on-chain and refreshes it when the price moves enough to matter. A pull oracle holds a signed value off-chain and posts it only when an application asks, which saves gas and cuts delay for markets that reprice every few seconds.