Web3 Security Weekly News Review — August 31–September 6, 2026

This week's review covers August 31 – September 6, 2026: 13 publicly reported events. The week is the largest of the year in dollar terms — roughly $331.6M gross — because the Liquid Network, a federated Bitcoin sidechain, was exploited via a rangeproof verification cache-key collision: 3,998.5 unbacked L-BTC (about $320M) were minted without any peg-in and converted to real BTC within minutes.

Two facts reshape that headline. Around 3,400 BTC (~$272M) were returned to the federation peg wallet after the fix, and roughly 598.5 BTC (~$48M) remain attacker-held with an on-chain bounty negotiation ongoing. On a net basis the week is ~$59.6M — still the largest of the site's coverage. Excluding Liquid entirely, protocol losses were ~$11.6M, led by the Injective exchange-module collision.

What was the biggest Web3 security story of Aug 31 – Sep 6, 2026?

The Liquid Network exploit of September 6: an attacker minted ~3,998.5 L-BTC (about $320M) with no corresponding peg-in by exploiting a cache-key collision in Elements' rangeproof verification, then pegged out to real Bitcoin within minutes — draining roughly 95% of the federation's reserves. About 3,400 BTC were returned after the patch; ~598.5 BTC remain with the attacker.

Overview

The week's shape is one enormous event plus a busy protocol tail. The Liquid Network incident is the largest publicly disclosed security event involving a Bitcoin sidechain this year, and its mechanics matter beyond the dollar figure: the flaw was a performance optimization — caching verified rangeproofs — whose cache keys were built from concatenated variable-length fields without length prefixes. Distinct arguments could collide into the same key, letting an invalid proof inherit a legitimate one's 'verified' verdict. The attacker seeded caches with crafted setup transactions, then minted 3,998.5 L-BTC with no backing and pegged out within minutes; the federation reserve fell from ~4,200 BTC to ~197 BTC.

Around that event, the protocol tail ran ~$11.6M: the Injective binary-options insurance-fund collision (~$4.8M, two defects lining up in the chain's own exchange module), the Aquifer Solana AMM drain (~$2.5M, credential-compromise suspected), Notional's escrow claim forgery (~$1.73M, straight into Tornado Cash), and the XRPH Wallet keyspace collapse that swept 4,011 wallets (~$452K) — a wallet vendor whose 55-character entropy string reduced key generation to a 2^46 search space. Enforcement was present: Tether's September 2 blacklist froze ~$2.39M across 25 addresses, 23 of them on TRON.

The 13 Stories

#01Sep 6Bridge / ConsensusHIGHConfirmed

Liquid Network rangeproof cache collision ($320M gross; ~$272M returned)

An attacker exploited a cache-key collision in the Elements codebase's rangeproof verification — cache keys concatenated variable-length fields without length prefixes, so distinct parameter sets could reuse a cached 'verification passed' result. Two crafted setup transactions seeded node caches; a third minted 3,998.5 L-BTC with no peg-in. The unbacked L-BTC was consolidated, spent and pegged out to Bitcoin mainnet within minutes, taking the federation reserve from ~4,200 BTC to ~197 BTC (~$320M at ~$80K/BTC). Peg operations were suspended, the sidechain paused, and Elements v23.3.4 (with proper length prefixes) shipped; the network restarted September 10. Roughly 3,400 BTC (~$272M) were returned to the peg wallet after the fix; ~598.5 BTC (~$48M) remain attacker-held, with the attacker demanding a 10% bounty via OP_RETURN and Blockstream refusing a ransom.

My take: The federation keys were never compromised — a speed optimization from 2016 was. Cache keys are security boundaries wherever 'already verified' is a cached verdict, and every project with a similar pattern should audit it this week.

Source: SlowMist analysis (Sep 11); CoinDesk; The Block; SideSwap and Liquid Federation statements

#02Aug 31Exchange ModuleHIGHConfirmed

Injective binary-options insurance-fund collision (~$4.8M)

Two defects aligned in Injective's exchange module: an insurance fund's identifier collided with a binary-options market's identifier (fields concatenated without separators), binding an INJ-denominated fund to a USDC-quoted market — and the settlement path never checked that a fund holds its market's denomination. The attacker traded against their own subaccounts to manufacture a 12,744 USDC shortfall, which the protocol covered from an INJ balance worth a fraction of a cent, then withdrew far more than deposited — ~$4.8M in USDC.

My take: Both the identifier construction and the missing denomination check were written by the same team on the same path. Composition review inside a single codebase is still adversarial review.

Source: BlockSec incident analysis

#03Aug 31ProtocolMEDConfirmed

Aquifer Solana AMM drained (~$2.5M)

Solana-based proprietary AMM Aquifer was exploited for about $2.5M. Public reporting points to compromised protocol-linked wallets or admin credentials rather than a confirmed smart-contract bug; the attacker operated addresses on both Solana and Ethereum. The project posted an on-chain white-hat offer: return at least 80% of assets by September 3, 14:00 UTC and keep up to 20%.

My take: The white-hat offer with a deadline is becoming standard practice — the next update that matters is whether the deadline was met.

Source: SlowMist; Aquifer on-chain statements

#04Sep 3–4ProtocolMEDConfirmed

Notional Finance escrow claim forgery (~$1.73M)

An attacker submitted a setup transaction late on September 3 and drained an escrow contract in the first minutes of September 4 (block 25,900,234): 69,257.38 DAI and 1,658,524.86 USDC, about $1.73M. The apparent mechanism was a fraudulent fCash claim settled as legitimate — an unsafe-cast class error in how the escrow validated claims. The stolen stablecoins were swapped into 689.2 ETH and deposited into Tornado Cash.

My take: The protocol had the correct check — on a different conversion in the same function. The funds went straight to a mixer, so recovery odds are the worst of the week's cases.

Source: PeckShield; BlockSec; Notional incident communications

#05Aug 31ProtocolMEDConfirmed

Ankr FLOW unbacked mint into More Markets (~$410K)

An attacker exploited a vulnerability in Ankr's ankrFLOW liquid-staking contract to mint ~8.6M unbacked ankrFLOW tokens, then used them as collateral with More Markets' E-mode to drain ~15.5M WFLOW (~$410K at spot) from the lending reserve; realized proceeds were ~$246K after slippage. The Flow Foundation attributed the root cause to Ankr's Solidity contract, not Flow EVM or More Markets.

My take: The unbacked-asset-into-lending pattern shows up again — the pause modifier guarded one staking entry point and left its sibling open.

Source: SlowMist; Flow Foundation statement

#06Aug 31Access ControlMEDConfirmed

Radix access-control exploit (~$1.2M)

Radix was hit by an improper access-control exploit with estimated losses of ~$1.2M.

My take: Access control stays on the monthly board; authorization paths repay review more often than they get it.

Source: DefiLlama hacks database

#07Sep 3–4WalletMEDConfirmed

XRPH Wallet keyspace collapse — 4,011 wallets swept (~$452K)

XRPH Wallet's credential generation passed a 55-character value into xrpl.Wallet.fromEntropy(), which expects raw bytes — only the first 16 characters survived, cutting the keyspace from 2^128 to roughly 2^46 (72.9 trillion combinations), with Math.random() potentially weakening it further. A mass sweep across 10,281 payments from 4,011 sender wallets (Sept 3–4) moved 267,664 XRP plus 23.2M XRPH and 2.43M XRPHAI — roughly $450–452K — into a collector account, later traced to 445,198 DAI on an Ethereum address. The team reproduced keys for nine live wallets using public data. XRP Healthcare announced a wind-down on September 10.

My take: No malware, no phishing, no XRPL flaw — just entropy done wrong once for every user of one app. Wallet vendors should publish key-generation entropy tests the way they publish audits.

Source: CryptoSlate; xrpl.to analytics; XRP Healthcare statements

#08Sep 2EnforcementLOWConfirmed

Tether blacklist wave — $2.39M frozen, 23 of 25 addresses on TRON

Tether blacklisted 25 addresses on September 2, freezing approximately $2.39M in USDT — $2.24M across 23 TRON addresses and $145K on Ethereum. The largest single freeze held ~$1.39M.

My take: Same-day issuer freezing remains the most effective recovery lever in the stablecoin layer, and TRON remains the dominant venue by address count.

Source: USDT blacklist tracking (TRONSCAN-derived)

#09Aug 31ProtocolLOWConfirmed

Balancer V1 rounding/precision exploit (~$234K)

An attacker exploited a rounding and precision vulnerability in legacy Balancer V1 contracts, using flash loans to compress WBTC reserves to near-zero, minting BPT with one satoshi of WBTC, and proportionally exiting with the pool's DPI, USDC, WETH and WBTC — approximately $234K.

My take: Legacy pools live on shared front ends: the bug was old, the reach was new.

Source: SlowMist

#10Sep 5ProtocolLOWConfirmed

Secured Finance spot-price manipulation (~$104K)

Secured Finance's lending protocol was exploited via spot-price manipulation for approximately $104K on Ethereum.

My take: The spot-price dependency list keeps growing; every name on it is an unforced error waiting for liquidity to thin.

Source: DefiLlama hacks database

#11Sep 5ProtocolLOWConfirmed

Dream Health Chain reward-logic flaw (~$72K)

Dream Health Chain on BSC was exploited through a reward-logic flaw for approximately $72K.

My take: Reward accounting remains the most common small-protocol failure mode.

Source: DefiLlama hacks database

#12Aug 31ProtocolLOWConfirmed

Float Protocol share mispricing (~$28K)

A flash loan manipulated a Uniswap V3 pool's spot price (slot0), mispricing Float Protocol's Hypervisor LP shares; repeated deposits and withdrawals at inflated values extracted ~$28K (10.71 ETH). No TWAP or slippage protection guarded the critical functions.

My take: The smallest loss of the week carries the week's most repeated lesson: spot price as truth.

Source: SlowMist

#13Sep 5ProtocolLOWConfirmed

Reddio RedSonic share-accounting exploit (~$23K)

Reddio's RedSonic vault on Ethereum was exploited via incorrect share accounting for approximately $23K.

My take: Share accounting closes the week as it opened at Injective: when a protocol computes its own value, the computation is the attack surface.

Source: DefiLlama hacks database

Key Signals

SignalReading
Gross losses~$331.6M — the largest week of the year, on one event
Net losses~$59.6M after ~$272M of on-chain returns (Liquid)
Top vectorVerification and accounting trust: cache collision, identifier collision, forged claims
ContainmentLiquid paused 4 days, patched (v23.3.4), restarted Sep 10; ~85% of the event returned
TRON activityTether blacklist Sep 2: $2.24M frozen across 23 TRON addresses
Biggest watch itemWhether similar cache-key patterns exist in other verification systems

TRON Footprint

ItemOn-chain footprintStatus
Tether blacklist wave, Sep 2 (~$2.39M)USDT blacklist applied on-chain; 23 of 25 addresses on TRON; visible on TRONSCANConfirmed

For Tron energy users: the blacklist mechanism is issuer-level and works without touching your address's other assets, but the practical takeaway is the same as ever — USDT received from unknown counterparties can become tainted, and tracing runs on public data. Energy costs for a routine transfer remain trivial next to one contaminated deposit.

What I'm Watching Next Week

  • The Liquid Network restart (completed September 10) and the on-chain bounty negotiation for the remaining ~598.5 BTC.
  • SlowMist's full Liquid analysis (published September 11) and whether other Elements-based or cache-using systems audit the same pattern.
  • The Aquifer white-hat deadline (September 3) — whether 80% came back.
  • Whether the XRPH keyspace collapse spreads to any other wallet built on the same code pattern.

Sources

  1. SlowMist, Liquid Network incident analysis (September 11) and Hacked database entries.
  2. CoinDesk and The Block, Liquid Network coverage and federation statements.
  3. SideSwap incident statement and Liquid Federation notices.
  4. BlockSec weekly incident report, week of 2026/08/31 (Injective, Aquifer, Notional, Ankr).
  5. CryptoSlate / xrpl.to / XRP Healthcare, XRPH Wallet incident.
  6. PeckShieldAlert, Notional escrow tracing.
  7. DefiLlama hacks database (Radix, Secured Finance, Dream Health, RedSonic).
  8. USDT blacklist daily tracking, September 2 (TRONSCAN-derived).

This article is for general information only. It describes publicly reported security incidents; dollar figures are point-in-time estimates drawn from the sources listed above and may be revised as investigations continue. Nothing here is financial, legal or investment advice.

Frequently Asked Questions

Why is the Liquid Network event counted at $320M when most funds were returned?

$320M is what was minted unbacked and pegged out — the actual exposure on the day. Because ~3,400 BTC (~$272M) came back after the patch, I report the net figure (~$48M retained by the attacker) alongside the gross. Weekly totals are given on both bases.

Are these figures exact?

No. They are point-in-time estimates from the sources named on each card; the BTC-denominated figures in particular move with Bitcoin's price (~$80K at incident time). Event counts and share arithmetic are computed exactly from the list.

What is a rangeproof cache-key collision?

Elements caches verified range proofs to skip re-verification. The cache key was built by concatenating variable-length fields without length prefixes, so two different inputs could produce the same key. Crafted transactions made an invalid proof collide with a legitimate one's cached verdict — nodes then accepted an unbacked output.

Were any TRON users affected this week?

Yes — the September 2 Tether blacklist froze ~$2.24M across 23 TRON addresses. That is enforcement, not a loss to TRON users; no confirmed TRON-user losses were reported in the window.

Sources & Methodology

Sources

  1. Chainalysis - on-chain threat intelligence & loss data.
  2. SlowMist - security monitoring and incident advisories.
  3. CertiK & PeckShield - smart-contract incident analyses.
  4. TRONSCAN & main block explorers - on-chain verification.
  5. Project official blogs / X / GitHub - incident statements.
  6. Mainstream industry media, cross-checked against the whitelist.
  7. Figures labeled 'unconfirmed / under investigation' where the trail is not public.

This article is for general information only. It describes publicly reported security incidents; dollar figures are point-in-time estimates drawn from the sources listed above and may be revised as investigations continue. Nothing here is financial, legal or investment advice.

Strengthening your defenses? See how transaction fees and energy costs scale on Tron at Tronsell.io.