Selfish mining attacks are a strategy in which a miner deliberately keeps newly found blocks secret instead of publishing them right away. By releasing those hidden blocks at carefully chosen moments, the miner can cause honest miners to waste their effort and earn a larger share of rewards than its computing power would normally justify.
The idea challenged a comfortable belief about proof of work: that a blockchain stays fair as long as no single party controls a majority of mining power. This article explains how selfish mining works, why it can be profitable, what signals might reveal it, and how protocol designers and pool operators respond.
Why a Minority Miner Might Break the Rules
In proof of work, miners repeatedly hash block candidates until one produces a value below the network’s difficulty target. The first to succeed publishes the block and earns the reward. Honest miners announce a block the instant they find it.
The assumption behind this behavior is that announcing quickly is always the best choice. If you wait, someone else may find a competing block, and your work could be wasted. For years, many people treated honest publishing as automatically incentive compatible.
Academic research by Ittay Eyal and Emin Gün Sirer argued otherwise. Their work showed that a well-connected miner with substantial but minority hash power could sometimes gain more by withholding blocks. Mining fairness, it turned out, depends on more than just majority control.
The finding prompted extensive follow-up research into mining incentives, and the term selfish mining became standard vocabulary in blockchain security discussions.
The Core Mechanics of Block Withholding
The strategy revolves around maintaining a secret lead over the public chain and revealing it to cancel honest work. Two phases describe the attacker’s behavior.
Building a Private Lead
When the selfish miner finds a block, it keeps that block private and continues mining on top of it. Honest miners, unaware of the hidden block, keep working on the older public tip.
If the attacker finds another block before the public network does, its private lead grows to two blocks. Every block honest miners produce on the public chain during this time is at risk, because the attacker holds a longer chain ready to publish whenever it chooses.
Releasing Blocks to Orphan Honest Work
When honest miners find a block and close the gap, the attacker publishes enough private blocks to stay ahead or match them. Nodes following the longest valid chain switch to the attacker’s version.
The honest block becomes orphaned, also called stale: valid but abandoned, earning nothing. If the attacker’s lead was only one block, a race follows in which part of the network may build on each version. The outcome depends heavily on how quickly the attacker’s block reaches other miners.
Why the Strategy Can Pay Off
Selfish mining does not create extra blocks. In the short term, the attacker loses some blocks of its own whenever a race goes badly. The profit comes from the share of rewards, not their absolute number.
By forcing honest miners to waste work on blocks that end up orphaned, the attacker reduces the effective output of the rest of the network. The attacker’s surviving blocks then form a larger fraction of the main chain than its share of raw hash power.
Over time, many proof of work networks adjust difficulty based on block production speed. Once difficulty reflects the reduced effective output of honest miners, the attacker’s relative gain can turn into a higher absolute reward per unit of time. Analysts continue to debate the exact conditions under which this effect appears.
A Hypothetical Race Between Two Pools
Consider a pool called Pool S that finds a block at the same height honest miners are still working on. Instead of announcing it, Pool S keeps mining privately on top of its hidden block, hoping to extend its lead before anyone else catches up.
An honest pool, Pool H, then finds its own block and broadcasts it. Pool S immediately releases its hidden block. Nodes now see two valid blocks at the same height, and the network splits as miners build on whichever version arrived first.
If the next block lands on Pool S’s branch, Pool H’s block is orphaned and its reward vanishes. If it lands on Pool H’s branch, Pool S loses instead. The attacker accepts these gambles because, over many rounds and with good connectivity, the losses fall disproportionately on honest miners.
Real pools do not announce such behavior, which is why the strategy is studied mostly through models, simulations, and statistical analysis of fork data.
The Role of Network Connectivity
A key variable in selfish mining analysis is how fast the attacker’s blocks propagate compared with honest blocks. During a tie, the miner whose block reaches more of the network first tends to win.
An attacker with excellent connectivity, many peers, and low-latency links to other mining pools can win more of these races. In theoretical models, stronger connectivity lowers the share of hash power needed for selfish mining to be profitable, sometimes to well below a majority.
This is why selfish mining sits at the intersection of consensus and networking. Weak peer-to-peer propagation does not cause the attack, but it can amplify an attacker’s advantage. Network-level threats such as eclipse attacks, where a node is isolated from honest peers, can make the strategy easier to execute.
Effects on the Network and Its Users
Even when it falls short of a majority attack, selfish mining harms the system in several ways. The damage affects miners, users, and the long-term structure of the network.
- Wasted energy: honest miners spend resources on blocks that are later orphaned.
- Centralization pressure: if selfish pools earn more, rational miners may join them, concentrating hash power.
- Less reliable confirmations: more frequent reorganizations make shallow confirmations less trustworthy for merchants.
- Higher majority risk: a pool that grows through selfish mining moves closer to the level that enables double spending.
The centralization effect is often considered the most serious. A strategy that rewards size encourages consolidation, and consolidation undermines the decentralization that makes proof of work secure. Selfish mining and majority attacks are therefore connected risks rather than separate problems.
Smaller networks feel these effects more sharply, because a single pool can represent a large share of total hash power.
Signals That Might Reveal Selfish Mining
Detecting selfish mining with certainty is difficult because individual events look like normal network noise. Natural forks happen when two honest miners find blocks at nearly the same time.
Patterns observed over long periods can raise suspicion. Unusually high orphan rates, frequent short reorganizations, or one pool repeatedly releasing several blocks in rapid succession may indicate withholding. Block timestamps clustered unnaturally or blocks arriving just after rival blocks are also worth noting.
Researchers have proposed statistical tests that compare observed fork patterns with what honest mining would produce. Such evidence is circumstantial, though. Network monitors should collect data from many geographically spread nodes, since a single observer can misjudge propagation timing and wrongly accuse a pool.
Public data on orphaned blocks is incomplete, because nodes discard stale blocks they never received. Independent observers sharing their logs give a fuller picture than any single monitor.
How Selfish Mining Differs From a Majority Attack
A majority attack, often called a 51 percent attack, involves controlling most of the network’s hash power. Such an attacker can reliably outpace the honest chain, reverse confirmed payments, and censor transactions.
Selfish mining is subtler. The attacker does not need majority power and usually does not aim to reverse specific payments. Its goal is economic: earning an outsized share of rewards by making honest work less productive.
The two can interact. A selfish miner who grows its share through higher rewards may eventually approach majority control. A majority attacker may also use withholding tactics while preparing a reorganization. For broader coverage of attacks on agreement rules, see our overview of consensus layer attacks.
Recognizing which threat is present helps decide the right response, from adjusting confirmation counts to encouraging miners to move to other pools.
Proposed Defenses in Protocol Design
Researchers and developers have suggested several changes to reduce the profitability of withholding blocks. Each involves trade-offs, and not every proposal has been adopted by major networks.
- Random tie-breaking: when two competing blocks of equal height arrive, nodes choose randomly instead of keeping the first seen, reducing the edge from fast propagation.
- Timestamp-based rules: fork choice considers block freshness, penalizing blocks that appear to have been held back.
- Rewarding uncle blocks: some designs credit valid but orphaned blocks, lowering the cost honest miners suffer from forks.
- Faster block relay: improved propagation protocols shrink the time window in which races occur.
Each idea addresses one ingredient of the attack. Random tie-breaking limits the connectivity advantage, while better relay reduces natural forks that hide withholding. Designers must check that a defense does not introduce new manipulation vectors, such as falsified timestamps.
What Mining Pools and Miners Can Do
Individual miners usually join pools, which combine hash power and share rewards. The pool operator decides which block template members work on and when to publish found blocks.
Miners should choose pools that publish clear policies, share data about found blocks, and operate transparently. A pool reporting unusually high returns relative to its hash rate without explanation deserves scrutiny. Pool operators, in turn, can publish block discovery times and propagation logs to demonstrate honest behavior.
Spreading hash power across many independent pools reduces the risk that any single operator gains enough share for selfish mining to pay. Some protocols allow individual miners to select their own transactions within a pool, which can also limit the operator’s control over block content.
Exchanges and payment services that track pool shares can also adjust confirmation requirements when concentration rises, adding a practical layer of protection for their customers.
Selfish Mining Beyond Proof of Work
Similar ideas appear in other consensus designs. Any system where a block producer can delay publication to gain advantage invites comparable analysis, even when the specifics differ.
In proof of stake networks, a validator chosen to propose a block might withhold it or time its release to influence fork choice. Protocol designers address this through attestation deadlines, proposer scoring adjustments, and penalties for missed duties. Behavior differs between networks, so readers should consult each protocol’s documentation.
The common thread is information asymmetry. Whenever one participant knows about valid blocks that others do not, it may gain leverage. Fast, open propagation and fork-choice rules that do not reward secrecy make this leverage smaller in every type of chain.
Studying these variants helps builders of new chains avoid repeating old incentive mistakes in a different form.
Watching for Withholding in the Years Ahead
Selfish mining is best understood as a reminder that incentives, not only cryptography, keep a blockchain honest. A chain can be technically sound and still reward behavior that slowly erodes decentralization.
Users and businesses can watch a few indicators: the distribution of hash power among pools, orphan and reorganization rates, and the transparency of large pools. Rising concentration or unexplained fork patterns are reasons to increase confirmation thresholds for valuable payments.
For developers, the lesson extends to any new protocol. Before launching, model how a rational participant with significant resources might deviate from expected behavior. Testing designs against strategic actors, not just honest ones, catches incentive problems before real value depends on them.
Sharing simulation results and assumptions publicly also invites outside reviewers to find weaknesses before attackers do.