Hyperliquid risk

TL;DR

Hyperliquid is its own Layer-1: HyperCore runs the perp DEX, HyperEVM runs contracts, and USDC is now mostly minted natively on the chain rather than escrowed on Arbitrum. The venue had a very good month. Core open interest went from $6.99B to $12.79B, HYPE set a new all-time high of $94.48 earlier today, HLP's yield came back off the floor, and a native borrow and lend market opened on 2026-09-18. Nothing broke. No exploit, no bridge incident, no emergency validator action between 2026-08-05 and 2026-09-19.

The verdict stays caution, and the reason is that none of the three things holding it down moved.

Validator concentration is the first. Five Hyper Foundation validators hold 47.96% of stake and the top three alone hold 36.69%, still above the one-third that can block anything needing a two-thirds quorum. Those numbers drifted down by about a point. A point is not a structural change.

HIP-3 is the second. Outside teams still deploy perp markets on Hyperliquid's book and margin engine and still set their own oracle prices. The dollar exposure is flat at $3.83B. What changed is the denominator: because core open interest nearly doubled, HIP-3 fell from 36% of the platform to 23%. That is a real improvement in the shape of the platform and no improvement at all in the mechanism. The docs still say a slashed deployer's stake is burned rather than paid to affected users, and the manipulation review still only triggers when a price moves more than 50% against the start of day, which is more than twice the move that liquidated $57.4M in July.

The third is new to this report and it is partly a correction of our own last one. The audits page names one firm, Zellic, on the "legacy" bridge contract. That bridge now holds 8.6% of the collateral tagged to it. The rest sits as natively minted USDC on a chain whose consensus, matching engine and clearinghouse have no named external audit, and a credit layer took $269M of borrows on its first day with no audit named either. We rated this yellow on 2026-08-05 and we should not have.

Checklist

Audits and contracts. The docs page reads "The Hyperliquid legacy bridge contract has been audited by Zellic," with two reports, plus a pointer to Circle's separately audited hyperevm-circle-contracts. One word carries a lot of weight there. The USDC page is explicit: "USDC is natively minted on the Hyperliquid L1" and "The legacy Arbitrum bridge holds less than 10% of the USDC supply on HyperCore," with the Zellic audit covering "the legacy bridge and its logic in relation to the L1 staking." DeFiLlama's own chain split agrees on the direction: of the $6.94B tagged to the bridge child, $6.345B sits on Hyperliquid L1 and $598M on Arbitrum, so Arbitrum is 8.6%. Note that this share has been rising, not falling: 5.1% on 2026-09-01, 7.5% on 2026-09-18, 8.6% today. So the single named audit governs under a tenth of the collateral, and no external audit is named anywhere for HyperBFT, HyperCore, HyperEVM, or the borrow and lend system that opened on 2026-09-18. Two honest qualifiers. This is an absence of a named audit, not evidence of a bug, and the empirical record over three years at very large scale is clean. And most of this was already true when we rated it yellow last month, so treat the color change as us fixing our own reading rather than as something breaking.

Admin control. Pulled live from the validator API. 27 active validators, 8 inactive, 442,012,355 HYPE staked, worth about $41.2B at the live mark. Five Hyper Foundation validators hold 211,978,499 HYPE, which is 47.96%, down from 48.91% on 2026-08-05. The top three, all foundation, hold 36.69% between them, down from 37.45% and still above the one-third blocking level. Eight validators together reach 68.33%. Six validators show as jailed and all six are already inactive: three of them (Kraken, Falcon Capital, Figment) hold 10,000 HYPE each and the other three hold under 105 HYPE, so together they are under 0.01% of stake. That is housekeeping, not an outage. The delisting power was exercised in the window: VINE was voted out around 09:30 UTC on 2026-08-26 and the API confirms it flagged delisted with zero open interest, which is what a completed force settlement looks like. That is scheduled exchange housekeeping, the same category as OM in February and IP in June, not an emergency. JELLY in March 2025 remains the emergency use and the one that shows what the power can do in two minutes.

Oracle. The mechanism is unchanged word for word. HIP-3 docs still assign the deployer "Market definition, including the oracle definition and contract specifications" and "Market operation, including setting oracle prices, leverage limits, and settling the market if needed." Slashing still runs by stake-weighted validator vote, up to 100% for irregular inputs causing invalid state transitions or prolonged downtime, up to 50% for irregular inputs causing brief downtime, up to 20% for invalid inputs degrading network performance, and the sentence that sets the color is still there: "the slashed stake by the deployer is burned instead of being distributed to affected users." The cross-margin section still puts the manipulation review at a 50% move in externalPerpPx against the start of day, with the added consequence that an asset moving that far more often than monthly becomes ineligible for cross-margin.

Correcting our own draft on one point, because it was wrong in a way that mattered. The perpDexs registry does not return a null oracle updater across the board. Four of the ten builder dexes carry an explicit non-null oracleUpdater key, and two of them, Felix and EntropyIO, share the exact same one. Where the field is null, as it is for Trade.xyz, the deployer key itself holds oracle authority. Either shape is a single key. The finding from last month therefore survives in substance and is now specific: price authority on builder markets resolves to individual keys, some of which are reused across separate dexes.

Two other things changed and they point in opposite directions. The registry now lists ten builder dexes rather than nine, with a new one, EntropyIO, at $56.3M across eight markets, mostly AI-company names. And the docs carry a backstop liquidator: each HIP-3 dex gets a fully onchain strategy at 0x400..00 plus the dex index that takes over backstop-liquidatable positions in cross-margin assets, falling back to ADL to guarantee solvency. This was not in our 2026-08-05 report, though we cannot date when the docs added it. Read it precisely. It reduces the chance of auto-deleveraging during volatility. It does nothing about a bad oracle print, because a liquidation executed on a wrong price is a correct liquidation as far as the engine is concerned.

The size picture is genuinely better. HIP-3 carries $3.83B against $12.79B on core, so 23% of the $16.62B platform, down from 36%. Trade.xyz is still 97.9% of all HIP-3 open interest at $3.75B over 108 markets, and xyz:SKHX, the market that broke in July, now carries $339.7M, slightly more than the $333M it held last month. A second SK-prefixed market on the same dex, xyz:SKHY, carries $194.4M; we did not verify this run whether it tracks the same underlying. Concentration inside HIP-3 is close to total, and six of the ten registered dexes have no live markets at all today.

Liquidity and exit. This is the dimension that improved most. Core open interest is $12.79B across 178 live markets: BTC $3,578M, ETH $2,968M, HYPE $2,039M, ZEC $854M, SOL $657M. Discount some of that, because open interest is quoted in dollars and HYPE went from $57 to $93 over the same period, so part of the growth is the mark and not new positions. Even discounted it is the deepest perp book in the sector by a wide margin. The exit path also changed shape. Instead of an Arbitrum bridge that needs two-thirds stake-weighted signatures for every withdrawal, most USDC is natively minted and moves by Circle CCTP to any supported chain. That removes the bridge as the single custody chokepoint, which is a real improvement, and replaces it with reliance on L1 correctness plus Circle. DeFiLlama's series dates the flip precisely: Arbitrum held 62.6% of this collateral on 2026-06-01, 14.6% by mid-June and 5.7% by 2026-07-01. Which means our 2026-08-05 report describing the Arbitrum bridge as the custody surface was already a month behind reality when we published it, and the 7.1% figure that proved it was sitting in our own appendix.

Yield: real versus emission. Character unchanged, size much better. HLP and staking pay from funding and trading fees, and the new borrow market pays from borrower interest with the protocol keeping 10% as a liquidation buffer. Nothing here is inflationary. HLP's own API now reports an APR of 3.70%, up from 0.34% last month, with all-time PnL of +$138.16M against +$136.9M on 2026-08-05. That delta, +$1.26M, is the number that settles the question, and it is the one to use rather than the API's 30-day PnL, because the API's month window opens on 2026-08-19 and misses two weeks of the stretch. The vault itself kept shrinking, from $216.7M on 2026-08-05 to $187.5M now, about 13.5%, and with PnL positive across the whole period the cause is redemptions rather than losses. Median predicted staking APR across the active set is 2.16%, essentially flat on last month. A vault earning real fees that depositors keep leaving is not a solvency problem and is not a green either.

Holder concentration. Genesis split has not changed: 31% community distribution, 23.8% core contributors, 6% Hyper Foundation budget, 38.888% future emissions, 0.3% grants, 0.012% HIP-2, with no private-investor, CEX or market-maker allocation. Core-contributor tokens mostly vest across 2027 and 2028. The distinction from last month still holds and is worth repeating because the headline number invites the wrong read: foundation-operated validators hold 211,978,499 HYPE, which is 47.96% of stake and 21.2% of the 1,000,000,000 max supply, against a foundation genesis budget of 60M. Most of that stake is delegated by other people. It counts fully for voting power and not at all for ownership, and delegators can move.

One thing to add this month, because it changes how any concentration ratio should be read. Total staked HYPE, 442M, is about twice CoinGecko's reported circulating supply of 222.4M. That is not a contradiction; locked and vesting tokens can be staked while being excluded from circulating supply. It does mean two things. Ratios computed against circulating supply for this token are meaningless, and the foundation validators' 212M of voting stake is close to the size of the entire tradeable float. Wallet-level concentration for that float is still not measured. The Nansen top-holders call returned HTTP 403 this run because the endpoint requires a Pro subscription, so this is a tooling gap rather than a data gap, and retrying with a different address will not fix it.

Track record. Clean at the core and still carrying one hosted failure. DeFiLlama's hacks list has exactly one entry attached to the Hyperliquid protocol id, $37,000 on 2023-06-15, classified as market manipulation via risk parameter abuse. Two things around that deserve stating rather than eliding. First, a trap: the list also carries an entry dated 2026-08-06 named "Hyperliquid Malaysia," and that one is a Solana wallet key compromise with target type "Wallet," no amount, and no DeFiLlama protocol id. It does not attach to this protocol and should not be read as a Hyperliquid incident. Second, apps hosted on Hyperliquid L1 have been exploited even though the core venue has not: LND for $1.18M in May 2025, HyperVault for $3.6M and Hyperdrive HL Lending for $782K in September 2025, Raga Finance for $18.5K in November 2025, Purrlend for $1.5M in April 2026. None is Hyperliquid itself and none falls in this window, but the chain is not a place where nothing has ever gone wrong. Separately, on 2026-08-24 a single Hyperliquid user lost $550,000 in USDC to a fake search advertisement tied to the Inferno drainer, with no protocol or infrastructure involvement. The July SK Hynix cascade stays in the record exactly as we wrote it: $57.4M liquidated across 960 accounts on a Trade.xyz HIP-3 market, roughly $17.4M of realized losses repaid voluntarily by the deployer, and no protocol mechanism that could have done the same. Whether the reimbursement completed after the 2026-08-15 claim deadline is not confirmed this run.

Worst case

Three now, and they still do not share a mitigation.

The HIP-3 oracle case is the one that already happened and the one the protocol has no answer to. You hold a position on a builder-deployed market, the deployer's feed ingests a real but unrepresentative print, and you are liquidated at a price the asset never traded in size. Burning the deployer's 500,000 HYPE, now worth about $46.6M, pays you nothing by explicit design, and the manipulation review does not even open below a 50% daily move. In July a deployer chose to pay and called it one-time. Assume the next one does not. The mitigation is free and total: check the ticker prefix and stay on core markets. The surface is still $3.83B, and there is a second version of it that costs nothing to trigger. A HIP-3 deployer can settle an asset with haltTrading, which cancels all orders and settles positions to the current mark price. HyENA did exactly that across its whole book between 2026-08-31 and 2026-09-02, delisting one market per hour after $4B of lifetime volume, because Hyperliquid's move to USDC left less room for its USDe-margin model. Orderly, announced, and entirely at the deployer's discretion. Your position closes on their schedule, not yours.

The validator case is unchanged. A large coordinated position pushes HLP into deep loss, the set intervenes, the market is delisted, and your open trade settles at the one-hour time-weighted spot oracle price before the vote rather than at the market. No insurance fund covers a directional HLP drawdown or a settlement you disagree with.

The new one is a credit case, one day old at the time of writing. HYPE posts as collateral at 65% LTV and BTC at 50%, with partial liquidation at (1 + LTV) / 2, so 82.5% for HYPE and 75% for BTC. That means a HYPE-collateralized borrower is liquidated on a roughly 21% drawdown from the point where they max out. $269M was borrowed on day one, a figure attributed to co-founder Jeff Yan rather than measured. A leverage layer priced against HYPE, sitting on the venue where HYPE's own perp carries $2.04B of open interest, is reflexive by construction: a sharp HYPE drawdown now forces collateral sales on the same chain that prices it. It is too young to have a track record, no external audit is named for it, and the global and per-account caps live on a docs page this run did not read.

Bottom line

Caution, for the third month, and it is worth being clear that this is not the same caution as a protocol going sideways. The core venue is in the best shape we have measured it in: $12.79B of core open interest, the deepest book in the sector, a working native-USDC path that removed the Arbitrum bridge as the custody chokepoint, and yield that comes from fees rather than emissions.

Three dimensions are red. Admin control, because a 27-validator set with the foundation at 47.96% and three of its own validators past the one-third blocking level has already voted a market out of existence in two minutes, and delisted another one this month. Oracle, because HIP-3 still delegates price authority to third parties carrying $3.83B, that authority resolves to individual keys with at least one shared across two dexes, the penalty for abusing it burns rather than repays, and the one serious failure was repaired by a company's discretionary payment. Audits, because the only named external audit now covers a bridge holding 8.6% of the collateral tagged to it, and the credit layer that opened yesterday has none.

For our own book, nothing about the venue changes. We short HYPE on Pacifica against a long leg held through a Kinetiq-staked-HYPE Pendle PT, and we read funding off this venue. That funding comes from core validator-signed markets, not HIP-3 feeds, so it stays fine to trust. Two things to carry forward. First, a HYPE drawdown now has a new amplifier in the native borrow market, so size the long leg for a sharper move than last month's data implies, and note that the PT leg we reviewed on 2026-08-12 matures on 2026-09-24, which is five days out, so the rollover decision lands inside that window. Second, if anyone here is tempted by a builder-deployed market because the yield or the ticker looks interesting, the answer is still no.

Data appendix

• TVL: $7.12B (DeFiLlama parent slug hyperliquid, api.llama.fi/tvl/hyperliquid = 7,120,835,592 this run; the orchestrator's same-day pull returned 7,116,843,657, so the series ticks). The month-on-month figure needs a consistent basis and the draft of this report did not use one. Parent-only, DeFiLlama's own series reads $6.042B on 2026-08-05 against $7.122B today, about +17.9%. On the parent-plus-HLP basis our last report used, $6.259B against $7.309B, about +16.8%. Either way the month is roughly +17%, not the +13.5% that comparing today's parent against last month's parent-plus-HLP would produce. TVL is still the wrong headline for a venue like this; open interest is nearly double it.
• TVL basis, demonstrated: DeFiLlama lists five children under parent#hyperliquid (Hyperliquid, Hyperliquid Bridge, Hyperliquid HLP, Hyperliquid Spot Orderbook, Hyperliquid Perps), two of which report no separate TVL. Live today: hyperliquid-bridge $6,943,391,927, hyperliquid-hlp $187,434,276, hyperliquid-spot-orderbook $177,443,665. The parent equals bridge plus spot-orderbook to the cent, and excludes HLP because HLP sits inside the bridge. Summing all three overshoots the parent by exactly the HLP line. Do not sum the children.
• Custody split: parent TVL is Hyperliquid L1 $6,522.1M plus Arbitrum $599.8M. For the bridge child specifically, Hyperliquid L1 $6,345.2M and Arbitrum $598.2M, so Arbitrum is 8.6%. DeFiLlama's history for that child: Arbitrum was 71.5% on 2026-05-01, 62.6% on 2026-06-01, 14.6% on 2026-06-15, 5.7% on 2026-07-01, 7.1% on 2026-08-05, 5.1% on 2026-09-01, 7.5% on 2026-09-18. The migration to native USDC therefore completed in June 2026, before our previous report, not this month. The share is currently rising, not falling.
• Bridge and USDC: Hyperliquid docs, USDC page, re-read this run: "USDC is natively minted on the Hyperliquid L1," Circle's HyperEVM contracts at 0xb88339cb7199b77e23db6e890353e22632ba630f, Circle CCTP as the Arbitrum-to-HyperCore route, and "The legacy Arbitrum bridge holds less than 10% of the USDC supply on HyperCore. The legacy bridge and its logic in relation to the L1 staking have been audited by Zellic." Hyperliquid has publicly stated that "in the final state, the Arbitrum bridge will be deprecated and all USDC will be natively minted." No bridge hack to date.
• Open interest, core: $12.792B live (Hyperliquid API metaAndAssetCtxs, 2026-09-19). BTC $3,578M, ETH $2,968M, HYPE $2,039M, ZEC $854M, SOL $657M. 178 live markets out of 234 universe entries; 56 are flagged isDelisted and every one carries zero open interest. Up from $6.99B and 177 live markets on 2026-08-05, about +83%, part of which is the HYPE and majors mark rather than new positions. The delisted count went 55 to 56 and live went 177 to 178, so one delisting and two new listings since 2026-08-05.
• Open interest, HIP-3: $3.825B across the live builder dexes, 23.0% of the $16.617B platform total, down from 36% on 2026-08-05 while the dollar figure stayed roughly flat ($3.89B then). Trade.xyz (xyz) is $3,745.0M over 108 live markets of 123, which is 97.9% of HIP-3 open interest; largest are xyz:SP500 $421.8M, xyz:SKHX $339.7M, xyz:GOLD $296.5M, xyz:SKHY $194.4M, xyz:XYZ100 $182.5M. EntropyIO (io) is $56.3M over 8 live markets (io:ANTH $34.9M, io:NBIS $9.7M, io:OAI $7.3M). Paragon (para) $16.5M over 27, Markets By Kinetiq (mkts) $7.4M over 4. Six registry entries return no live markets at all: flx Felix Exchange, vntl Ventuals, hyna HyENA, km Markets by Kinetiq, abcd ABCDEx and cash dreamcash. Note km and mkts share the deployer 0x71f0019cc7fa79e4f42587fb7b9a817d8d2429ec.
• HIP-3 registry and oracle keys (corrected this run): ten builder-deployed dexes plus core, up from nine on 2026-08-05. The perpDexs response does not return a null oracle updater for every entry, contrary to this report's draft. Four carry an explicit oracleUpdater: flx at 0x94757f8dcb4bf73b850195660e959d1105cfedd5, io at the same address, hyna at 0xaab93501e78f5105e265a1eafda10ce6530de17e, and para at 0x8888888c43cbb7e1c4132542e46831bffd866ed3, which is also its deployer. The field is null for xyz, vntl, km, abcd, cash and mkts, which means the deployer key itself holds oracle authority. Trade.xyz's deployer is 0x88806a71d74ad0a510b350545c9ae490912f0888. The substance of last month's finding holds: builder-market price authority resolves to single keys, and at least one key is shared across two registered dexes.
• HIP-3 rules (docs re-read 2026-09-19): deployer owns "the oracle definition and contract specifications" and "setting oracle prices, leverage limits, and settling the market if needed." Stake 500,000 HYPE, about $46.6M at the live mark, minimum 183 days. Slashing up to 100% for irregular inputs causing invalid state transitions or prolonged downtime, up to 50% for irregular inputs causing brief downtime, up to 20% for invalid inputs degrading network performance, amount set by stake-weighted median of validator votes. "The slashed stake by the deployer is burned instead of being distributed to affected users." Manipulation review only each time the externalPerpPx of an asset "moves more than 50% relative to the start of day price," and an asset exceeding that more often than monthly becomes ineligible for cross-margin and subject to slashing. Settlement: the deployer may settle an asset using the haltTrading action, which cancels all orders and settles positions to the current mark price.
• HIP-3 backstop liquidator (not in our 2026-08-05 report): each HIP-3 dex is associated with a fully onchain strategy at "0x400..00 + {dex_index}" that takes over backstop-liquidatable positions, accepting only cross-margin-enabled assets and falling back to ADL to guarantee solvency. This reduces ADL during volatility. It is not a remedy for a bad oracle price. We cannot date when the docs added it.
• HIP-3 permissioned markets: co-founder Jeffrey Yan announced HIP-3*, an opt-in permissioning add-on, on 2026-09-03, letting a deployer keep an onchain allowlist of wallets or delegate that to an approved sub-deployer. Available in preliminary form on testnet, optional, existing markets unchanged, mainnet pending a future network upgrade. Not on mainnet this run.
• Native borrow and lend (new, live 2026-09-18): docs FAQ "Manual borrows." Borrow USDC or USDT against supplied HYPE (65% LTV) and BTC (50% LTV). Supplied HYPE and BTC do not earn interest; supplied USDC and USDT earn interest but do not contribute to borrow capacity. Partial liquidation threshold is (1 + LTV) / 2, so 82.5% for HYPE and 75% for BTC. The protocol "retains 10% of borrowed interest as a buffer for future liquidations." Interest accrues continuously and is indexed hourly, with borrows subject to available liquidity and to account and global caps documented elsewhere. Roughly $269M borrowed on day one, attributed to co-founder Jeff Yan via press (KuCoin, CryptoBriefing, CryptoSlate), not verified against an API pull. No external audit is named for this system on the docs audits page, and the FAQ does not name the oracle that prices the collateral.
• Validator set: 27 active, 8 inactive (Hyperliquid API validatorSummaries, 2026-09-19). Total active stake 442,012,355 HYPE, about $41.2B at the live mark. Five Hyper Foundation validators hold 211,978,499 HYPE, 47.96% (was 48.91%): HF3 12.56%, HF2 12.25%, HF1 11.88%, HF4 7.82%, HF5 3.45%. Top three, all foundation, 36.69% (was 37.45%), still above the one-third that blocks a two-thirds quorum. Eight validators reach 68.33%. Six validators show isJailed, all of them inactive; Kraken, Falcon Capital and Figment hold 10,000 HYPE each and HyperCN X hlscan, Red Pond and cp0x by STAKR.space hold under 105 each, so under 0.01% of stake between them. Median predicted staking APR (month period) 2.1637%, against 2.178% last month. Validator commissions range 0% to 10%.
• HLP vault: $187.47M (Hyperliquid API vaultDetails, 2026-09-19), cross-checked against DeFiLlama hyperliquid-hlp at $187.43M the same day. Down from $216.7M on 2026-08-05, about 13.5%. Reported APR 3.70% (apr 0.0369855), up from 0.34% last month. All-time PnL +$138,158,268 against +$136.9M last month, so +$1.26M over the 45 days, which is the right window and confirms the decline is redemptions and not losses. The API's own 30-day PnL of +$1.02M opens on 2026-08-19 and does not cover the stretch; do not use it for the month-on-month comparison.
• Audits: Hyperliquid docs audits page, re-read 2026-09-19. Names Zellic only: "The Hyperliquid legacy bridge contract has been audited by Zellic," with two reports linked, plus "Circle's contracts are audited independently" linking circlefin/hyperevm-circle-contracts. No external audit named for HyperBFT, HyperCore, HyperEVM, or the borrow and lend system. A bug bounty program page exists separately. Absence of a named audit is confirmed; the existence of an unpublished one is not ruled out.
• Admin and governance: validators can delist assets and force-settle. Perps "settle to the 1 hour time weighted spot oracle price before the scheduled delisting voting time," and on delisting "all positions are settled and open orders are cancelled." Inside this window, VINE was delisted following a validator vote at about 09:30 UTC on 2026-08-26; the API confirms VINE flagged delisted with zero open interest, as are FTT, IP, OM and JELLY. Delisting votes have run fully onchain with automatic execution on quorum since the March 2025 upgrade that followed JELLY. No change to the bridge validator set or withdrawal controls was found in the window.
• HYPE price and funding: mark $93.219, oracle $93.179, prior day $91.11, 24h notional volume $428.1M, hourly funding 0.0000125, which annualizes to 10.95% (Hyperliquid API, 2026-09-19). CoinGecko the same day: $92.99, all-time high $94.48 set 2026-09-19 at 03:05 UTC, market cap $20.69B, fully diluted $88.84B, +26.1% over 30 days. Press reported a $92.56 print on 2026-09-18 at the lending launch, above a prior high of $89.57. Up from $57.215 on 2026-08-05, about +63%. These tick continuously; treat as a snapshot.
• Token distribution: 1,000,000,000 HYPE max supply, CoinGecko total supply 955,307,079 and circulating supply 222,445,714. Genesis 31% community distribution, 23.8% core contributors, 6% Hyper Foundation budget, 38.888% future emissions and rewards, 0.3% community grants, 0.012% HIP-2. No private-investor, CEX or market-maker allocation. Core-contributor tokens locked one year post-genesis, most vesting 2027 and 2028. Foundation-operated validators control 21.2% of max supply as stake, the large majority of it delegated by others rather than owned. Read the float carefully: total staked HYPE (442M) is about twice reported circulating supply (222.4M), because locked tokens can be staked, so concentration ratios computed against circulating supply for this token are meaningless. Wallet-level holder concentration: refresh (the Nansen top-holders endpoint returned HTTP 403, premium labels require a Pro subscription).
• Recent-news and incident scan (2026-08-05 to 2026-09-19): no exploit, no bridge incident, no emergency validator action, no HIP-3 oracle failure or liquidation cascade, no HLP loss event, no governance change, no chain halt or outage. An X-wide scan over the window returned only routine operational chatter. DeFiLlama's hacks list holds exactly one entry attached to the Hyperliquid protocol id, $37,000 on 2023-06-15, market manipulation via risk parameter abuse. Attribution warning: the same list carries an entry dated 2026-08-06 named "Hyperliquid Malaysia," chain Solana, target type "Wallet," classification key compromise, with no amount and no DeFiLlama protocol id. It is not a Hyperliquid protocol incident and should not be counted as one. For completeness, the list does carry exploits of other protocols hosted on Hyperliquid L1, none in this window and none of them Hyperliquid itself: LND $1.18M (2025-05-09), HyperVault $3.6M (2025-09-26), Hyperdrive HL Lending $782K (2025-09-27), Raga Finance $18.5K (2025-11-22), Purrlend $1.5M (2026-04-25). Separately, on 2026-08-24 a single Hyperliquid user lost $550,000 in USDC to a counterfeit search advertisement tied to the Inferno drainer, with no protocol or infrastructure involvement. Material non-incident events in the window: the native borrow and lend launch (2026-09-18), the HIP-3* permissioned-markets testnet proposal (2026-09-03), the VINE delisting (2026-08-26), and HyENA winding its markets down one per hour between 2026-08-31 and 2026-09-02 after $4B of lifetime volume across 12,000 users, driven by Hyperliquid's USDC alignment rather than by any failure.
• Open items carried from July: whether Trade.xyz completed SK Hynix reimbursement after the 2026-08-15 claim deadline is not confirmed this run. refresh.

Maintained monthly. Methodology: DeFi Research Instruction v2.