Reports & audits
Dated analyses,
not rumours.
Every report is sourced, states its limits and is published under responsible disclosure. Reports are ranked by severity, with the most severe marked in red.
Archive
Reports by severity.
KMD and Conceal remain the two critical alerts, shown with red cards. Cosmos, S-7, S-8 and Ledger are rated medium: they require access to the local file or a precondition.
Komodo / AtomicDEX macOS: critical entropy
Weak RNG, compromised wallets, BTC moved and a KMD balance still exposed. Public alert and limits of the analysis.
Conceal: DNS checkpoints and bypassed consensus
Checkpoints are loaded over plain DNS and, inside their zone, PoW and signatures are no longer verified. A DNS attack can make a forged chain accepted.
Cosmos: 9 chains, zero direct path
Audit of 9 Cosmos/CometBFT chains. No direct path in the surfaces examined; Kava P1, Akash/Sei P2, Celestia R3/R4.
S-7 · NavCoin NEXT: 10/10 historical versions
10/10 states of server-lw.js (2019–2021) are affected: the AES key is derived by MD5 of the password. Access to the local file is still required.
S-8 · Epic Cash: 13/13 tags + master
13/13 tags + master use PBKDF2-HMAC-SHA512 with 100 iterations. Access to wallet.seed is still required for the offline attack.
LTE implant in resold Ledger devices
Supply chain, hardware implant and eSIM exfiltration. Public analysis and limits of interpretation.
No reports in this category yet.
How to read this rating. Severity reflects observed impact, ease of exploitation and affected scope. A report may be reclassified as new evidence emerges. Times are Paris time (CEST). To report an incident: security@frblockchain.icu.
Full library
The other 28 reports.
The 6 detail cards are above; the 28 cards below are public summaries. Full technical reports remain restricted.
Ripe Protocol
All 7 funding surfaces were checked one by one and ruled out; no permissionless direct theft or minting was found. Main risks: the Robinhood timelock is 0 (requires 2 multisig keys), StabVault absorbs unowned balances, PSM conditional arbitrage, and Addys._getAddys performs zero validation (no loss so far).
Ledgity Yield
On Base, lyEUR book NAV deviates from real backing by about 58.6% (2.41×); there is currently no secondary market and deposits are at the same price → no permissionless arbitrage. The redemption cap of 113,056 units (117,367 EURC) matches the aEURC buffer pool wei for wei, and the provider payment route has been disproven; the risk is holders front-running before the owner corrects NAV. On the permission surface there are also F-11 (CCIP 2-of-2 can mint) and F-12 (on Hedera, the entire chain reduces to 1 EOA).
Lucid Labs
No permissionless direct theft or minting was found; F-1 disables multi-bridge insurance (minBridges=0), so a single bridge failure leaves 100% of that pool (about $213k) exposed; also rescueTokens does not exclude principal, and LToken.mint skips the cap for admin.
Vetro Protocol
0 permissionless profit paths; the two Treasuries' KEEPER_ROLE holders total 5 pure EOAs (3 each, 1 overlapping), and for a single Treasury only 1 private key is needed to permanently freeze all redemptions; if the third-party Hemi vault fails, redemptions halt.
Umbra Privacy V2
Not a single key: upgrade authority for the 9 programs sits on the vault PDA of the same Squads multisig, 2-of-5 with time_lock = 0 → 2 member keys can take over everything with zero delay (F-1); the pool holds about 722.250822 SOL (measured 2026-11-01; originally measured 994.82) plus 46,712 USDC. A permissionless third party cannot take it: the 107 pool token accounts have clean authority (delegate/closeAuthority both 0), SOL sits in program-owned accounts, and escape destinations are bound by proof; the issuer's Token-2022 extensions only affect dust; the handler logic cannot be proven without source, but that does not change asset control.
Mendi Finance
Byte-identical to upstream Compound (9 market implementations match); no permissionless theft; the pricing layer has zero staleness checks and that layer has never been audited; pauseGuardian is a pure EOA that can freeze liquidations and transfers indefinitely.
Hyperbeat USD
No permissionless minting; the exchange rate only goes up and losses cannot be booked → after a loss, whoever exits first takes real assets at an inflated rate. §4.3/4.4 were additionally completed: 8 Morpho markets (LLTV 62.5%, utilization 86–93%), with liquidation pricing routed through the RedStone oracle that has no staleness checks (F-4).
Monolith Market
High: after the feed stops updating, a position can be liquidated within the 24h linear price-decay window, and dust is enough to take the entire collateral; any permissionless user can trigger it; 5 markets, TVL about $697K.
Flex
R1/R2 disproven; R4 not disproven, with 1 further medium-severity finding F2 (upfront fee uses the post-insertion average rate); the Dutch auction starts at the oracle price with a 99% floor and deactivates once it breaks below; this is not a zero-audit target (4 audits).
Kinza Finance
No permissionless theft or minting (all 25 reserves are ACL-guarded and upgrades sit behind a 4h timelock); RISK_ADMIN is an ordinary EOA that can change eMode priceSource with no timelock (about $1,670 to lever $1.21M of WBNB reserves).
Threshold thUSD
No permissionless minting; 200.6 million supply vs 538,000 system debt (SP deposits 49.99 million); each of the two PCVs has debtToPay of 100 million and an actual balance of about 50.04 million each.
DexFi / Monetrix batch
The DexFi factory owner is 2-of-3 with no timelock (1 signer is an EIP-7702 delegated account) and can swap the beacon to replace 1,901 vaults at once; the vault implementation returns 404 on all three paths and cannot be audited. All Monetrix minting entry points are controlled.
Balancer v2
Confirmed Critical rounding flaw (two mulDown sites, same root cause as the 2025-11-03 incident): no oracle required, zero capital, completed in a single batchSwap; this entry is a post-mortem, the incident already occurred (about $125M / 98% of TVL) and it is not treated as a current target.
Symbiotic Core
No theft; deallocatable() has zero fault tolerance, and a single-point revert freezes the entire exit path (the LiquidLane adapter has no try/catch wrapper).
Morpho Vault V2
Structurally conservative with no price-inflation cash-out entry point; N1: _totalAssets=0 is an irreversible absorbing state; N2: permissionless profit skimming during the post-loss recovery period (requires a loss first).
AimBot
The dividend contract holds 67.458 ETH; a single EOA calling updateDividends() can lock it permanently; phantom shares 1,082,024 vs tokens 1,000,000 (+8.2% dilution); the question of whether a third party can take the funds was exhaustively assessed as no.
BananaGun
Exploited on 2024-09-19 ($3M / 11 users); the root cause was the off-chain Telegram message oracle and could not be discovered on-chain; on-chain contract balances are only 0.004–0.23 ETH.
ENKI / eMetis
F-1 Critical: a single EOA holds upgrade authority + minting authority + Config, and can instantly take all METIS in the redemption queue; F-2 High: minter_mint has no cap + addMinter has no delay. Both are key-holder permission surfaces.
Qubic contracts
All 32 contracts were covered; issueAsset has only 4 call sites and each was read line by line (2 are fixed-supply/private, 2 can only issue to self and require payment); of 244 writable entry points, 213 move funds and 109 carry caller==state authorization; no permissionless minting or theft.
UniBot Router
There is no evidence that the 2023 access-control flaw was fixed; however there are no victims at present (the router holds zero assets and the top 50 holders have allowances of 0); there is also 1 unlimited USDC allowance sitting on a zero-balance address.
VaultCraft V2
Black box (three contracts unverified, no public repo): the accounting is closed and nobody can be drained; but OracleVaultController.owner() is a pure EOA that can change prices, set limits, and pause.
ZAMM
Pool liquidity cannot be drained (K() constant product + transient balances are physically backed); F-1 real fund loss: _payOut never pays the maker, so funds are stranded (currently about 0.024 ETH).
Splice
All 4 markets are isExpired (2024-08-21); stranded funds exist but belong only to PY holders or the treasury and cannot be taken by third parties (it would be Critical only if beneficiary were written as msg.sender).
Tranche / Fence
Fence TranchePool (2026-10-28) lost 1,013,353.61 USDC; the borrowing side is fully permissionless, anyone can open their own facility.
Smallcap sweep
Owner/proxy state scan of AirPuff, VaultCraft, Loop, and ENKI (not a full audit): for AirPuff's 15 upgradeable contracts, all ProxyAdmin owners are pure EOAs.
Pirate + Epic
Pirate has no HIGH/CRITICAL; LOW: -rpcauth is not implemented, a strcpy stack overflow at komodo_utils.cpp:421-424, -rest is unauthenticated, and a 3.9 MB precompiled wget64.exe was committed.
Unisat Wallet
No weak entropy source was found; the extension vault uses only 10,000 PBKDF2 iterations, and the bitcoinjs patch disables 4 PSBT checks. No web wallet exists (none of the 86 routes is /wallet, and none of the 113 chunks contains bip39/keystore) → this attack surface does not exist on the web side; mobile APK: Math.random used 0 times, native CSPRNG, KDF via WebCryptoVaultEncryptor (600k).
Multi-chain aggregate report: 55 targets
§0–§61; about 55 targets / 50+ repositories (including prototypes, testnets, unaudited and already-excluded items); about 30 reviewed at code level; 0 instances of permissionless direct theft or minting; §61 closing re-check (Umbra conclusion, Hyperbeat F-4, both Unisat surfaces, Ledgity provider disproven); this is not 55 production mainnets.
34 reports total: 6 public detail cards + 28 public summaries. Cards do not link to third-party sites; full reports are provided only to authorized parties.