OAK — OnChain Attack Knowledge

Worked example · 2026-04

Hyperbridge Merkle-proof forgery and counterfeit-DOT mint — Polkadot ⇄ Ethereum / Base / BNB / Arbitrum — 2026-04-13

Loss
approximately $2.5M realised (per Crypto Briefing updated estimate, ~10× the initial $237K Defi News / The Block / Cointelegraph figure once the full asset scope was reconciled). Notional damage: ~$1.2 billion in counterfeit bridged-DOT tokens minted on Ethereum (1 billion DOT) plus comparable counterfeit mints across ARGN (~$1B notional), MANTA, and CERE on Ethereum, Base, BNB Chain, and Arbitrum. The notional-versus-realised gap (~$1.2B notional, ~$2.5M realised) is structurally informative — the counterfeit-mint primitive scales infinitely on the supply side but is bounded on the cash-out side by destination-pool liquidity. The price of native DOT fell ~4% ($1.22 → $1.18) on the news.
OAK Techniques observed
OAK-T6.005 — Proxy-Upgrade Malicious Switching. OAK-T6.006 — Counterfeit Token Impersonation. OAK-T10.002 (Message Verification Bypass — the MMR-proof-verification flaw allowed an outdated valid proof to be accepted as authorisation for a forged cross-chain message, bypassing the bridge's message-verification logic). OAK-T10 (Bridge / Cross-Chain parent).
Attribution
pseudonymous attacker. The on-chain attacker EOA is observable; no public attribution to a named operator group at v0.1. Hyperbridge's team paused the protocol post-event and shipped a patch within days. The exploit was confirmed publicly by Polkadot's official forum within hours of detection. Forensic coverage by The Defiant, Crypto Briefing, The Block, DL News, Cointelegraph, AMBCrypto, CoinDesk, CoinMarketCap (CertiK research), and CryptoTicker triangulates the technical root-cause to the MMR-proof reuse / forgery flaw.
Key teaching point
The Hyperbridge case is the canonical non-key-compromise proxy-upgrade-malicious-switching worked example for OAK. Unlike Steadefi (August 2023, internal-key compromise) where the attacker reached the upgrade authority via stolen keys, the Hyperbridge attacker reached the equivalent admin authority via a forged cryptographic proof that the bridge accepted as legitimate. The upgrade itself was the attack vector; the bridge admin role was changed in-flight via a counterfeit message. T6.005 contributors should treat key-compromise-driven and proof-forgery-driven proxy-upgrade-malicious-switching as separate sub-classes with different detection signals (off-chain-key-handling-monitoring vs on-chain-message-verification-correctness) and different mitigation surfaces.

Summary

Hyperbridge launched on Polkadot in November 2024 as a cross-chain bridge designed to enable trust-minimised message-passing between Polkadot and EVM chains via Merkle Mountain Range (MMR) proof verification. The Ethereum-side gateway contract was the load-bearing surface: it accepted MMR proofs over Polkadot consensus, verified them against committed state roots, and applied the verified messages to the Ethereum-side bridge state — including admin operations on the bridged-token contracts.

On 2026-04-13, an attacker exploited a flaw in the MMR proof verification logic that allowed an old, valid cryptographic proof from an earlier legitimate transaction to be accepted as authorisation for a different message. By pairing the outdated proof with a carefully crafted message that masqueraded as an authorised admin operation on the Ethereum-side DOT token contract, the attacker:

  1. Re-assigned admin authority on the bridged-DOT token contract to an attacker-controlled address.
  2. Used the acquired admin authority to mint 1 billion counterfeit bridged-DOT on Ethereum without depositing any backing DOT on Polkadot.
  3. Repeated the primitive against bridged-ARGN (~$1 billion notional minted), bridged-MANTA, and bridged-CERE across Ethereum, Base, BNB Chain, and Arbitrum.
  4. Cash-out via Uniswap into ETH; due to liquidity constraints in the destination pools, the realised cash-out was ~$237K initially (per same-day reporting) and ultimately revised to ~$2.5M as the full asset scope was reconciled (per Crypto Briefing updated estimate).

Polkadot's official forum confirmed the exploit within hours; Hyperbridge paused the protocol and shipped a patch over the following days. The native DOT token fell ~4% on the news; the counterfeit bridged-DOT was effectively worthless once the bridge was paused and the destination pools were drained.

The case is OAK's canonical T6.005 + T6.006 worked example because the upgrade-itself-as-attack-vector primitive is on full display: no operator keys were stolen, no pre-deployment social engineering occurred, and the bridge contract was the legitimate, audited deployment up to the moment the forged proof was accepted. The attacker's leverage was the on-chain message-verification logic, and the resulting admin-authority transfer was indistinguishable from a legitimate admin-rotation message at the contract layer.

Why this is structurally significant

T6.005 and T6.006 sit at an unusually clean intersection in this case: the proxy-upgrade-malicious-switching primitive (T6.005) was the mechanism by which the counterfeit-token-impersonation surface (T6.006) was opened. The two Techniques are structurally distinct but operationally chained.

Three structural features distinguish the Hyperbridge case within the broader proxy-upgrade-attack cohort:

  1. The upgrade authority was reached via on-chain message forgery, not via pre-stolen keys. Steadefi (August 2023) lost $1.14M when an internal-key compromise gave the attacker access to the deployer EOA, which then upgraded the proxy implementation to a malicious one. The attack vector was the key compromise; the upgrade was the downstream extraction primitive. Hyperbridge inverts this: the attacker had no operator keys, the deployer EOA was uncompromised, and yet the attacker achieved the equivalent upgrade-equivalent admin-authority transfer via a forged cryptographic proof that the bridge accepted as a legitimate admin-rotation message. The detection-signal locus differs entirely: Steadefi's failure mode lives at the off-chain key-handling layer (operator-side key custody monitoring); Hyperbridge's failure mode lives at the on-chain message-verification layer (proof-uniqueness-vs-replay correctness).

  2. The counterfeit-token leg is structurally different from typical bridge-token impersonation. The standard T6.006 sub-pattern is fake-symbol-matching deployment: a separate, unrelated contract is deployed with the symbol "USDC" or "WETH" and used in dust-attack lures or fake LP-token deposit drains. The Hyperbridge case is more severe: the counterfeit bridged-DOT was minted by the legitimate bridge contract, which made the counterfeit indistinguishable from legitimate bridged-DOT at the contract identity layer. Uniswap pools that quoted bridged-DOT would have priced the counterfeit identically to legitimate bridged-DOT — the impersonation succeeded at the protocol-identity layer, not just the symbol layer. This is the worst-case shape of T6.006.

  3. The notional-versus-realised loss gap is structurally informative. $1.2B notional minted, ~$2.5M realised cash-out. The ratio reflects the destination-pool-liquidity constraint: counterfeit mints scale infinitely on the supply side, but the operator can only convert as much to ETH / stablecoins as the destination pools can absorb without crashing the price. T6.006 contributors should record both notional and realised loss; the ratio is a useful structural signal across the cohort.

The case generalises beyond Hyperbridge. Any bridge whose admin authority on destination-chain token contracts is gated by cross-chain-message verification is exposed to T6.005 + T6.006 chained extraction if the message-verification primitive admits forgery, replay, or signature-aliasing. Wormhole's 2022 uninitialised-implementation bug was the EVM-side analogue (different mechanism, same downstream); Audius 2022 was the storage-collision EVM-side analogue. Hyperbridge is the cross-chain-proof-forgery analogue, and the worked example completes the v0.1 reference set for the proxy-upgrade-attack cohort.

Timeline (UTC)

When Event OAK ref
2024-11 Hyperbridge launches on Polkadot; Ethereum / Base / BNB Chain / Arbitrum gateway contracts deployed and operational (genesis surface)
Pre-2026-04-13 Bridge operates legitimately; multiple legitimate cross-chain admin / mint operations accumulate valid MMR proofs in the historical record (standing T6.005 surface)
2026-04-13 Attacker reuses an outdated valid MMR proof, paired with a forged admin-rotation message, against the Ethereum-side bridged-DOT token contract; admin authority transfers to attacker EOA T6.005 execution (proxy-upgrade-malicious-switching via proof forgery)
2026-04-13 (within minutes) Attacker uses acquired admin authority to mint 1 billion bridged-DOT on Ethereum; no backing DOT deposited on Polkadot T6.006 execution (counterfeit-token mint)
2026-04-13 (continuing) Attacker repeats against ARGN (~$1B notional), MANTA, CERE across Ethereum, Base, BNB Chain, Arbitrum T6.005 + T6.006 cohort scale-out
2026-04-13 (within hours) Attacker cash-out via Uniswap; ~$237K realised initially; pool liquidity constrains cash-out scale T5 outflow
2026-04-13 (same day) Polkadot's official forum confirms exploit on Hyperbridge's Ethereum gateway contract; Hyperbridge pauses the protocol; native DOT falls ~4% ($1.22 → $1.18) (defender response)
2026-04-14 onward The Defiant, Crypto Briefing, The Block, DL News, Cointelegraph, AMBCrypto, CoinDesk, CoinMarketCap (citing CertiK) publish forensic coverage triangulating the MMR-proof-reuse root cause (forensic surface)
2026-04 (revised estimate) Crypto Briefing updates realised loss to ~$2.5M (~10× initial figure) once the full counterfeit-asset scope is reconciled (loss-figure revision)
2026-04 onward Hyperbridge ships patch; protocol resumes operation under revised verification logic (recovery / mitigation)

What defenders observed

  • Pre-event (proof-verification layer): the MMR-proof verification logic accepted any valid historical proof against the claimed message rather than binding the proof uniquely to a specific message via per-message commitment. The flaw was a classic message-verification-correctness gap: the cryptographic proof was valid; the binding between proof and message was not enforced. This was a standing T6.005 surface from the moment the gateway contract was deployed in November 2024. Defender lesson: cryptographic-proof-verification correctness is binary at the contract layer; auditors and bridge designers must verify what the proof commits to, not just that the proof is cryptographically valid.
  • At-event (on-chain signal): the forged admin-rotation message produced an admin-transfer event on the bridged-DOT token contract from a non-deployer EOA, followed by mass-mint events with no corresponding deposit events on the Polkadot side. Either signal is detectable in real time by a bridge-monitoring indexer (admin-authority change without governance event; mint events without deposit-correlation). Hyperbridge did not have such monitoring in place at the time of the exploit; the public forum disclosure followed third-party detection rather than first-party telemetry.
  • At-event (price-surface signal): native DOT fell ~4% on the news, anchored at the moment Polkadot's forum confirmed the exploit. The price reaction was bounded because the counterfeit was confined to bridged-DOT-on-EVM; native DOT on Polkadot was unaffected. The price-surface is therefore a secondary signal; the primary signal is the on-chain admin-authority-change event.
  • Post-event: the counterfeit bridged-DOT was effectively worthless once the bridge was paused and Uniswap pool liquidity was withdrawn. The realised attacker profit (~$2.5M) is the upper-bound on cash-out within the available pool liquidity at the moment of execution. Operator-side response (Hyperbridge pause + patch) is the standard mitigation chain for a T6.005-class incident at the bridge layer.

What this example tells contributors writing future Technique pages

  • T6.005 has at least two distinct sub-classes. Steadefi (August 2023, internal-key-compromise sub-class — see citation list below for adjacent reference) and Hyperbridge (this case, on-chain-message-forgery sub-class) share a downstream extraction primitive (proxy admin authority used to drain or mint) but have different upstream attack vectors. T6.005 contributors should preserve the sub-class distinction: the detection signals diverge entirely between off-chain key-handling monitoring (Steadefi-class) and on-chain message-verification correctness (Hyperbridge-class).
  • T6.006 has at least three sub-classes. (a) Fake-symbol-matching deployment (separate unrelated contract sharing the symbol; canonical for dust-attack lures and fake LP-token deposit drains). (b) Bridge-impersonation via fake-bridge-token (a separate counterfeit deployment that masquerades as a legitimate bridged token; less severe than the Hyperbridge case). (c) Bridge-internal-mint (the legitimate bridge contract mints counterfeit tokens via attacker-acquired admin authority; the Hyperbridge case; the worst-case shape). T6.006 contributors should preserve the sub-class distinction.
  • The notional-versus-realised loss gap should be recorded in T6.006 cases. The ratio reflects destination-pool-liquidity constraints and is a useful structural signal across the cohort. Forensic coverage that reports only the realised cash-out figure underestimates the attempted extraction scale by 1-3 orders of magnitude in this class.
  • Cross-chain-proof-verification correctness is a v0.x-priority audit class. The Hyperbridge MMR-proof-reuse flaw is the canonical 2026 case; Wormhole's 2022 uninitialised-implementation bug and Audius's 2022 storage-collision are the EVM-internal analogues. The cross-cutting axis is contract-layer verification of authority-bearing messages; mitigation guidance should generalise across bridge and proxy patterns.
  • Detection at the bridge-monitoring layer is straightforward where deployed. Admin-authority-change events without governance events; mint events without deposit-correlation. The defender intervention is bridge-side telemetry, not contract-layer rewrite. Future T6.005 / T6.006 examples should document whether the affected bridge had pre-event monitoring in place.

Public references

Discussion

The Hyperbridge case is the canonical 2026 worked example for the upgrade-itself-as-attack-vector sub-class of T6.005 (Proxy-Upgrade Malicious Switching). The case is structurally cleaner than Steadefi (August 2023, internal-key-compromise sub-class) because no off-chain operator key was compromised — the entire attack chain lives at the on-chain message-verification layer. This makes Hyperbridge the right anchor for the contract-layer sub-class of T6.005, while Steadefi anchors the operator-key sub-class.

The case is also the canonical worked example for the bridge-internal-mint sub-class of T6.006 (Counterfeit Token Impersonation). The standard T6.006 pattern is fake-symbol-matching deployment — a separate, unrelated contract is deployed with the symbol "USDC" or "WETH" and used in dust-attack lures or fake LP-token deposit drains. The Hyperbridge sub-class is more severe because the counterfeit was minted by the legitimate bridge contract: at the protocol-identity layer, the counterfeit bridged-DOT was indistinguishable from legitimate bridged-DOT.

The triangulation across The Defiant, Crypto Briefing, The Block, DL News, Cointelegraph, AMBCrypto, CoinDesk, CoinMarketCap (CertiK), and AMBCrypto is unusually clean for a v0.1 case — multi-source coverage of the technical root cause (MMR-proof reuse), the realised-vs-notional loss accounting ($2.5M realised vs $1.2B notional), and the protocol-side response (pause + patch + Polkadot forum first-party disclosure). The case meets the v0.1 standard for "verified across at least 2 independent forensic providers" with substantial margin.

For OAK's broader cohort coverage, this case + the Solana PD burn-on-buy cohort (examples/2024-09-solana-permanent-delegate-burn-on-buy-cohort.md) + the cross-chain honeypot cohort (examples/2024-02-honeypot-token-cohort-cross-chain.md) collectively establish that the fake / honeypot / malicious-by-design smart contract cohort is a cross-chain pattern with structurally distinct sub-surfaces per chain ecosystem and per primitive (genesis-time hostile bytecode vs runtime-time hostile authority change vs cross-chain message-forgery). The cohort motivates the T1.006 / T1.007 / T6.005 / T6.006 sub-Technique proposals in TAXONOMY-GAPS.md.

Techniques demonstrated (4)