Skip to content
Crypto.world

Research

Inside Qubic's Selfish Mining Campaign on Monero: Evidence, Tactics, and Limits

In 2025 the blockchain network Qubic ran a mining campaign on Monero and publicly advertised it as a "51% takeover" and a demonstration of selfish mining — a strategy where a miner withholds blocks…

Inside Qubic's Selfish Mining Campaign on Monero: Evidence, Tactics, and Limits
arXiv
2512.01437
Published
2025-12-01
Authors
Suhyeon Lee, Hyeongyeong Kim

AI summary

Overview

Research area: Blockchain security and measurement — specifically proof-of-work mining incentives, selfish mining, and empirical network measurement of a live cryptocurrency (Monero).

Technical level: Intermediate. Readers need basic familiarity with proof-of-work, mining pools, orphaned blocks, and chain reorganization, but the paper's core arguments are presented largely in accounting and behavioral terms rather than heavy mathematics.

Scope: A measurement-and-modeling study of Qubic's 2025 mining campaign against Monero, combining node data, pool telemetry, community artifacts, and disclosed view keys to test Qubic's public claims of a "51% takeover" and selfish mining.

What This Paper Is About

In 2025 the blockchain network Qubic ran a mining campaign on Monero and publicly advertised it as a "51% takeover" and a demonstration of selfish mining — a strategy where a miner withholds blocks and releases them selectively to orphan competitors' work. Monero's privacy design makes this hard to verify: blocks carry no pool identifiers and withholding decisions happen off-chain, so public dashboards could not answer what actually occurred. The authors reconstruct the campaign from multiple independent vantage points to determine what happened on chain, how it was executed, and whether it actually paid off in mining rewards.

Key Contributions

  1. Multi-source attribution and dataset. The authors combine blocks observed by their own Monero node, Qubic pool job-notification signals, and community-collected Qubic network artifacts, then validate the resulting block labels against Qubic's disclosed view keys while explicitly quantifying coverage limits.

  2. Empirical campaign measurement. They reconstruct Qubic's mining activity, test the public claim of majority mining power, and measure the campaign's effects on orphaning and chain reorganizations rather than relying on hashrate estimates that cannot be measured directly.

  3. Strategy and economic evaluation. They compare observed behavior against the standard Eyal–Sirer selfish mining model and a new "conservative-release" variant (for which they derive a closed-form revenue expression, Proposition 1), analyze tie-breaking race outcomes, and account for delayed difficulty-adjustment spillovers when computing mining rewards.

  4. Operational response and inference limits. They document how community monitoring and Qubic's countermeasures changed what was observable over time, and treat this as a factor limiting what can be inferred about Qubic's intended release strategy.

Main Findings

  • Attribution is feasible despite Monero's privacy design. Community-observed data, extra-nonce patterns, pool observations, and disclosed view keys jointly support Qubic labels. Of 13,000 initially attributed Qubic blocks, 12,989 appear in the community dataset — a 99.92% confirmation rate — with no disagreement in chain or orphan status among matched blocks. The union contains 58,944 observed blocks, of which 13,782 are attributed to Qubic.

  • No sustained majority control. Qubic's observed block-production share averages 23.38% across the measurement period. Hourly series show short intervals approaching or briefly exceeding 50%, but daily and weekly aggregates never reach 51%, and no stable majority appears anywhere in the dataset.

  • Elevated orphaning and deeper reorganizations. Outside high-activity periods, orphan forks are almost exclusively length one; during high-activity periods the distribution shifts toward longer orphan chains and more frequent multi-block forks. Both Qubic and non-Qubic orphan counts rise markedly.

  • Wasted work from same-parent blocks. The authors directly verify eleven heights where at least two distinct Qubic blocks share the same parent: at nine heights one Qubic block entered the main chain and another was orphaned, and at two heights (in P2) both Qubic siblings were orphaned in favor of a non-Qubic block. Nine of the eleven cases occurred between August 14 and 17, including six in P2. Mining a second block on the same parent does not extend a private chain, so this is inconsistent with a rational strategy; the authors suspect an early job coordination failure.

  • Ten candidate selfish mining periods. A heuristic using orphan-count thresholds (τ_min = 2 orphan blocks per hour, d_min = 4 hours, g_max = 6 hours) yields periods P1–P10, where Qubic's estimated share averages 28.33% versus 23.38% over the full window.

  • Strategy shifted over time. In P1, P3, and P4, most runs providing release information correspond to release at lead 1 (standard selfish mining). In P8, the most intensive period, many runs align with release at lead 2, indicating a more conservative release rule. P2 contains four plotted Qubic self forks plus two additional orphaned same-parent pairs and provides no release information.

  • Neither static model matches all periods. At α = 28.33% and γ ≈ 0, the standard model predicts R_selfish ≈ 24.56% and the conservative model R_mod ≈ 23.71%. Qubic's observed P1–P10 main-chain share is 25.75% — below the honest baseline of 28.33%, and period points scatter between the two models, with P1 and P3 actually outperforming the honest baseline.

  • No aggregate reward advantage. Qubic's share of blocks expected at Monero's target rate is 25.11%, a 3.22 percentage-point shortfall from its estimated α. Without P8, its main-chain share is 29.88% against an estimated α of 31.53%. Transaction fees contributed below 3% of realized block rewards.

  • Local race wins exceeded period-level hashrate. Qubic's winning rate in resolved state 0′ tie-breaking events exceeded its period-level α in eight of ten periods — for example P1 at 0.49 against α = 0.35, P3 at 0.60 against α = 0.34, and P7 at 0.40 against α = 0.33. P2 recorded no Qubic-branch wins (0/14) and six parent-linked self forks within roughly 95 minutes.

  • Difficulty spillovers helped but did not close the gap. Across all gaps Qubic obtained 7,806 accepted blocks versus a replay estimate of 7,344.2 block-equivalents, a gain of +461.8 block-equivalents (+6.3%) — yet still 49.3 blocks below the target-rate expectation of 7,855.3. The P1–P10 period balance summed to −410.7 block-equivalents, and across periods plus gaps Qubic obtained 11,045 accepted blocks against 11,505.0 expected, a shortfall of 460.0 block-equivalents (4.0%). In the representative P8 case (α = 26.05%), Δ_active(P8) = −261.5 and Δ_DAA(G8) = +127.5.

  • Countermeasures reduced observability. The Monero community monitored Qubic's activity and relayed reconstructed blocks when possible; Qubic treated this exposure as an operational risk and progressively strengthened countermeasures, reducing visibility into its internal mining activity over time.

  • Disruptive incentives remain open. The authors explicitly restrict their profitability analysis to mining rewards and note that a "Goldfinger attack" — one whose utility comes from weakening the target cryptocurrency — could explain participation that block rewards do not.

Methodology in Plain English

The authors assembled three independent views of the same events. First, they ran a pruned Monero full node from September 29 to October 17, 2025, which let them see main-chain blocks plus locally observable orphan blocks that public explorers do not expose; for earlier periods they queried public nodes by RPC, probing 20 and selecting the five most responsive to improve orphan coverage. Second, they polled the publicly accessible Qubic mining pool's RPC API every 5 seconds from September 26 to October 17, 2025, using the job_notify method to see which block tip Qubic instructed its miners to build on. Third, they obtained community-collected Qubic block records from Monero community members, which added 93 orphan blocks absent from their node data.

To attribute blocks to Qubic, they started with an initial seed of orphan-fork blocks and refined two extra-nonce regex patterns, then cross-validated against the community dataset. Qubic's disclosed view keys — released only after each weekly epoch ended, so useless for real-time attribution — provided retrospective verification. To test timestamps, they measured the delay between each block's timestamp and the first subsequent job-fetch record referencing it, finding a mean of 5.61 seconds, with 86.3% within 8 seconds and 99.97% within 16 seconds, against an observed average fetch interval of 7.74 ± 1.46 seconds.

They then identified candidate selfish mining periods by counting orphans per hour and merging sustained stretches, and compared each period against two reference strategies: the standard Eyal–Sirer model and their conservative variant that fully publishes at lead 3. Finally, because slower block production can lower Monero's sliding-window difficulty, they replayed Monero's difficulty algorithm with synthetic slots inserted to restore target-rate production, estimating how much of Qubic's later gap revenue came from artificially cheap difficulty rather than from the strategy itself. Robustness was checked over all 80 parameter combinations of the period heuristic.

Why This Matters

This is one of the few documented, measured instances of a claimed selfish mining deployment on a live proof-of-work network, and it shows that a public "51% takeover" claim can be substantially weaker on chain than in marketing: no sustained majority, no reward advantage, but real disruption through orphaning and deeper reorganizations.

Real-world applications:

  • Pool operators and mining coordinators can use the same-parent block evidence as a diagnostic for job-coordination failures that silently waste hashrate.
  • Cryptocurrency communities and monitoring groups get a template for combining node observation, pool API telemetry, and community artifacts to attribute activity in privacy-preserving chains where blocks carry no pool identifiers.
  • Protocol designers can see how sliding-window difficulty adjustment transmits an attacker's disruption into later "cheap" blocks, which matters for evaluating intermittent-mining strategies.
  • Exchanges, custodians, and light clients relying on confirmation counts have a concrete reason to consider reorganization depth rather than orphan counts alone as a risk signal.

Industry relevance: the paper separates direct mining profitability from strategic disruption, which is the relevant distinction for anyone assessing whether an adversary is motivated by revenue or by damaging a target network. Its quantification of countermeasures reducing observability also matters for incident responders, who cannot assume their measurements stay reliable as an adversary adapts.

Future Directions

  • Distinguishing disruptive from profit-seeking motives. The authors leave open that incentives beyond direct mining rewards — such as Goldfinger-style utility from weakening Monero — could explain the campaign, and point to market-price evidence as a direction they only partially address.
  • Recovering internal attacker state. Because private-chain lead lengths and block-discovery times are unobservable, strategy inference rests on timestamps and orphan patterns; better observability or richer artifacts would allow stronger tests of the conservative-release model.
  • Closing coverage and timing gaps. The node data begin after the campaign was underway (September 29 versus an August campaign start) and job telemetry only covers September 26 to October 17, so earlier and short-lived activity may be missed; the authors also note that orphan coverage from public nodes is not guaranteed complete.
  • Evaluating mitigations and adaptive adversaries. The paper discusses pool-level detective mining and Publish-or-Perish-inspired defenses; whether such countermeasures work against an operator that progressively reduces its visibility remains an open question.

Target Audience

This paper suits blockchain security researchers and measurement specialists, mining pool operators and protocol developers, and analysts at exchanges or monitoring firms who need to assess reorganization risk. It is also valuable to Monero community members and incident responders interested in attribution under privacy constraints. Readers without basic proof-of-work and pool-architecture background will find the strategy discussion demanding, but the economic accounting sections are accessible to a motivated non-specialist.

Authors’ abstract

In 2025, the blockchain network Qubic conducted a mining campaign on Monero, a privacy-focused cryptocurrency. It publicly presented the campaign as a ``51\% takeover'' and a demonstration of selfish mining. The episode provides a rare opportunity to test such claims in a privacy-preserving proof-of-work network, where pool attribution and private release decisions are difficult to observe. We combine Monero node measurements, Qubic pool observations, community-shared…

The full abstract on arXiv
Read the original paper