Cross-chain security awareness in plain English. Not an audit. Not financial advice. We never publish drain recipes or exploit steps. News items are research snapshots from public reporting — not always a full 369 deep-dive.
16 Aug 2026: SafePal disclosed an authorization flaw in an order-tracking plugin. Under some conditions another customer’s order could be read. About 39,798 customers who ordered between 2 Mar 2025 and 11 Apr 2026 had name, email, shipping address, phone, and purchase details accessed externally. Vendor states seed phrases, private keys, wallet passwords, bank/card data, and government IDs were not in this set; cold storage is separate from the shop servers. Issue remediated; third-party review engaged; affected customers emailed from security@safepal.com. Real risk is targeted phishing (fake support, firmware, refund, delivery). Same class as the Trezor/ShipMonk shipping PII incident — not a protocol drain and not the Coldcard seed-entropy sweep.
369 take: A hardware wallet can still be “fine” while the shop that mailed it leaks enough PII for convincing fake-support plays. Never type a seed or backup into a site, app, or phone call that arrived after a vendor email. Check status only on the official SafePal security-update page. Pulse lesson: treat any wallet-vendor or logistics PII leak as phishing season for that cohort, not as proof the device itself was emptied.
18 Aug 2026 ~17:30 UTC: MAYAChain (Maya Protocol, a THOR-forked cross-chain DEX) halted after an attacker used chained Trade Account / outbound / pool-math bugs to inflate a thin pool (ARB.LINK cited) with a false subsidy, then extract about 48.87M CACAO and swap out. Attacker take is about $1.7M — majority ~20.8 BTC (~$1.34M class) still sitting in a PeckShield-cited Bitcoin address, plus other assets / on-chain CACAO. Team (Aaluxx) paused global ops. Pool value dropped ~$10.9M in the event, but most of that is CACAO crashing (~89% at the low) plus arbitrage — do not treat $11M as the amount stolen. SlowMist Hacked lists it 18 Aug. Not PulseChain. Not a THORChain 2026 vault replay.
369 take: A “theft protection” or subsidy path that can credit a pool when the reserve cannot actually pay is an unbacked mint in disguise. Separate what the attacker extracted from the later price crash. Pulse lesson: any AMM, teleport, or outbound-missing path that writes a new pool balance must fail closed if backing is not there — inherited THOR-style trade-account code still needs its own solvency check.
15 Aug 2026 23:32 UTC (BNB Chain block 116169049): DarkNavy-hosted researcher Defi Nerd flagged Fox Market on BSC. A one-transaction flash hit moved the Pancake V2 FOX/USDT pool while a bond/treasury mint still used the older pool price, over-issuing FOX/sFOX. Realized attacker take is about $119–120k (Defi Nerd ~$120k; Coinfomania $118.7k citing SlowMist). Larger damage is protocol overmint and LP impairment — we do not invent a USD total for that side. Confirmed success tx on BscScan matching the thread. This is BscScan token “Fox Market (FOX)” — not ShapeShift FOX and not the May 2026 FOX Colony drain on Arbitrum.
72h TI (2026-08-20) — SlowMist TI matches the 17 Aug card: 20 Aug 2026 digest: SlowMist Team TI (same ~$118.7–120k Fox Market BSC stale-spot stake/mint class) landed in the 72h Grok window. No new loss figure and no second card — this is corroboration of the existing BscScan + Defi Nerd item, not a new incident.
369 take: If mint, bond, or treasury size is taken from a live DEX pool that the same transaction can trade, the snapshot is attacker-controlled. Prefer a time-weighted or external price, a hard move cap, and no same-call buy-then-mint. Pulse lesson: same Allbridge / LpdFi-shaped gun on any PulseX pair used to price a mint.
15 Aug 2026: A close friend of the victim reported that a Coldcard Mk4 holder moved an entire ~$750k Bitcoin stack onto a well-known Australian exchange and that it was withdrawn in under 12 hours. Bitcoin News amplified the same thread: attackers had the Google account for about three months, including cloud-backed Google Authenticator, waited for the deposit, and the venue treated the withdrawal as the real owner; the victim reportedly saw an approval ping around 3 a.m. Verify: Unverified — friend-of-victim plus pickup, no named exchange, no explorer tx, no AU regulator or venue statement. The Authenticator class is independently real: 2023 cloud sync is not end-to-end encrypted, and as of March 2026 Google still had no E2EE update (Wirecutter). This is a CEX login / 2FA-cloud story, not the Coldcard seed-entropy sweep on the board.
369 take: A hardware wallet does not protect coins once they sit on an exchange whose login and 2FA live in the same Google account. Cloud-backed Authenticator is a spare set of codes for whoever holds that account — keep it offline if you use it, prefer a physical security key (two keys), and do not dump a cold stack onto a CEX because of a hardware scare. Pulse lesson: if your second factor syncs to the same cloud as your email, it is not a second factor.
30–31 Jul 2026: Coinkite warned that device-generated seeds on Coldcard Mk3 (firmware 4.0.1+) may put funds at risk. Updated advisory also flags seeds generated on Mk4 / Mk5 before 5.6.0 and Q before 1.5.0Q (~72 bits entropy vs expected 128); TAPSIGNER / OPENDIME / SATSCARD not in scope. Around the same window, ~594.48 BTC (~$38M) from ~500 single-sig wallets consolidated (Lookonchain cites bc1qnk4zh…). Causal link “sweep = this RNG bug” is not proven by Coinkite — treat viral “stolen due to Coldcard” lines as Real-overstated. Dice rolls (≥50 fair) and strong BIP-39 passphrases change exposure; migrate calmly per vendor guidance.
Update (2026-08-02) — Second sweep wave — Galaxy tracking ~1,158 BTC: 1–2 Aug 2026: Galaxy Research mapped a second wave attributed to the same operator class. Combined tracking ~1,158 BTC (~$70–75M class) from ~2,673 addresses across multiple attacker wallets (CoinDesk earlier expansion ~1,083 BTC / 1,196 wallets; some desks still climbing toward ~1,367 BTC). Firmware patch does not fix an already-weak seed — new seed + migrate. Causal “every sat = this RNG bug” remains research-attribution (Block/Galaxy), not a Coinkite courtroom proof — treat viral max $ as Real-overstated until settled. Enumeration is still running; don’t wait for a “final” loss total to migrate calmly.
Update (2026-08-02) — Third wave — +~208 BTC; Galaxy total ~1,367 BTC / 4,585 addresses: 1 Aug 2026 (Bitcoin Magazine pickup of Galaxy Research): a third sweep wave reported with another ~207.73 BTC moved. Combined tracking now ~1,367.05 BTC from ~4,585 addresses. Still the same enumeration campaign class — figures keep climbing as more bags are found. Review Coinkite’s official seed guidance; firmware alone does not repair an already-weak seed. Treat press “hack” shorthand and max-$ headlines with a Verify bar until causation vs total is settled.
Update (2026-08-03) — Fourth wave — ~449 BTC class; cumulative still climbing: 3 Aug 2026: Galaxy’s Alex Thorn flagged a likely fourth organized sweep while some spends were still in the mempool (~388–449 BTC range; later tally ~448.7 BTC from ~709 suspected addresses). RBF tip for unconfirmed spends only — once confirmed, migrate. Arithmetic add-on to prior three-wave ~1,367 BTC pushes observed totals toward ~1.8k BTC class — that sum is research tracking, not a Coinkite-confirmed final. Firmware update still does not fix an already-weak seed; generate new seed on fixed firmware and move calmly.
Update (2026-08-08) — Galaxy ~1,719 BTC (~$111 M) confirmed class; multi-attacker + early laundering: 7–9 Aug 2026 window: Galaxy Research high-confidence tracking cites ~1,719 BTC (~$111 M); wider desk estimates still range toward ~$130 M as enumeration continues. At least ~15 distinct attackers reported. Limited laundering observed (Wasabi-class BTC + Tornado Cash ETH). Coinkite migration guidance unchanged: fixed firmware exists, but updating alone does not repair an already-weak seed — generate a new seed on fixed firmware and migrate calmly. Treat viral max-$ totals as Real-overstated until settled; prefer Galaxy’s confirmed kernel for board language.
Update (2026-08-15) — Galaxy: ~1,779 BTC confirmed; sweeps look quiet after 6 Aug: 14 Aug 2026: Galaxy Research now puts high-confidence theft at 1,778.84 BTC (~$112.7 M) after speaking with 190+ victims, across more than 8,600 addresses. Confirmed attacker waves and footprints show no activity after 6 Aug — Galaxy reads that as most easy bags already moved or drained, not as “the bug is gone.” About 1,531 BTC still sits in attacker addresses; ~246 BTC has been moved (majority via CoinJoin). Firmware update still does not repair an already-weak seed — generate a new seed on fixed firmware and migrate calmly if you still hold on single-sig Coldcard addresses. A higher ~2,417 BTC / ~$153 M figure includes unconfirmed Wave 4 and other medium-confidence footprints — treat that as Real-overstated until Galaxy promotes it to the confirmed tally.
369 take: Seed quality and single-sig habits matter as much as smart-contract risk. Check device and firmware provenance, prefer strong passphrases or modern custody practices, and verify vendor advisories before you move size. Don’t treat on-chain aggregators as proof of root cause. Pulse lesson: long-dormant single-sig bags are a soft target when generation entropy was weak.
Volunteer Bitcoin Red Team (post–Coldcard) reports scanning ~501 open-source Bitcoin projects and filing ~7,958 security findings, with ~1,280 rated high or critical — numbers from team lead Calle’s campaign update (≈108 hours in), with earlier Bitcoin Magazine coverage of the first sprint (~390 repos / ~4,962 findings / 85 critical). OpenSats funds AI compute (tens of thousands of dollars); Kimi K3 has taken the majority of inference spend after early limits on some US lab models, with other models also used. CoinBureau’s “Kimi scanned nearly all of Bitcoin” framing is overstated: this is a human+AI campaign, findings are not all confirmed exploits, and only a minority had been reported upstream at last update. Maintainers are validating severity; patching is ongoing, not finished.
369 take: AI can flood a repo with “findings”; only maintainer-validated, reproducible issues matter for your bag. After Coldcard, the healthy response is more FOSS red-teaming — not panic that “China’s AI cracked Bitcoin.” Pulse lesson: treat viral AI-audit headlines as Verify-first; keep seed/RNG and wallet-library hygiene in the same bucket as protocol bugs.
12–13 Aug 2026: Scam Sniffer / Specter / PeckShield-class alerts mapped roughly $25.6–26M emptied from a related multi-wallet cluster in about 15 minutes (aWBTC, WBTC, DAI, ETH, aUSDC, LDO, sUSDe and more), then largely swapped into DAI and ETH. Firm reporting links this cluster to the same high-value holder who lost ~$24M to an increaseAllowance phishing event in September 2023. Current window is assessed as private-key compromise rather than another signature-phishing play — press notes at least one drained wallet had never granted token approvals. Dollar band confirmed across secondary press; exact wallet count varies slightly by writeup (multi-wallet cluster).
369 take: A prior phishing hit does not “clear” a wallet cluster — if keys or devices stayed in play, size remains a permanent target. Pulse lesson for holders and wallet UX: after any compromise, assume residual key risk, rotate, segregate, and never leave high balances on the same exposed set; monitoring related wallets beats hoping the first incident was the last.
13 Aug 2026: Trezor disclosed that ShipMonk, a fulfillment partner, had unauthorized access to customer order data. About 11,742 buyers had full exposure (name, email, phone, shipping address) and 1,947 partial (name, city, email) — limited to recent orders in US, UK, Sweden, Colombia, Brazil, Italy, and Portugal within the 90 days before 8 Aug 2026. Trezor states its own systems, devices, private keys, and wallet backups were not compromised. Affected customers were emailed from @trezor.io; the real risk called out is more sophisticated phishing by mail, phone, or email. Confirmed via official Trezor blog + @Trezor post.
369 take: A hardware wallet can be fine and you can still get hunted — name, address, and phone from a shipper are enough for fake “support / firmware / delivery” plays. Never enter a seed or backup on a website or over the phone; only trust updates on official Trezor channels. Pulse lesson: treat any vendor or logistics PII leak as a phishing season for that cohort, not as proof the device itself was drained.
10 Aug 2026: SlowMist logged ~70.83 ETH (~$136k class) lost from the USM protocol after a pricing flaw in redemption (`defund` / eth-from-defund style logic). The quote used an average of current and estimated final sell prices for a single redemption but was not split-invariant — combined with per-call state contraction and rounding, many small redemptions paid more ETH than one large redemption of the same size. Firm TI + secondary press; small dollar size, clean math lesson.
369 take: Redemption and mint math must pay the same whether the user splits the call or not. If small chunks beat one big call, attackers will loop it — often with a flash loan. Pulse lesson: stress-test vault / share / redeem paths for split invariance before size.
~11–12 Aug 2026: DefimonAlerts / Clara flagged an on-chain drain on Umbrae (Ignis AI Labs) DLMM pools on Base — roughly 25.7k USDC plus smaller WETH and cbBTC across several pools. Reporting describes unprivileged use of public pair / settlement-style functions to pull LP value. The team paused pools via a 3-of-4 Ops Safe and negotiated an ~80% white-hat return. Research snapshot from firm alert threads — not a full 369 deep-dive. Do not confuse with Pulse “Umbra” swap routers on the Risk board; this is a different Base DLMM product.
369 take: Public pair helpers and incomplete LP accounting are a live class even after audits. Pause authority and residual-balance checks matter when something goes wrong. Pulse lesson: any concentrated-liquidity or custom pair contract with callable settlement paths needs the same access and accounting review — and names that rhyme (Umbra vs Umbrae) are not the same product.
12 Aug 2026: Harmony said it is working with exchanges to stop and freeze funds after a suspected Layer-1 security incident, while preparing a patch and evaluating rollback options. On-chain analyst Juiceberg reported roughly 4 billion ONE minted without authorization (~26% of prior circulating supply class) with a large share routed toward exchanges; Harmony has not publicly confirmed the exact mint size. ONE sold off sharply (tens of percent). PeckShield-class alerts circulated in the same window. This is not the 2022 Horizon Bridge drain — treat it as an illicit mint / issuance-control failure while root cause details are still emerging. Exact billion-count remains research-attribution until Harmony confirms.
369 take: Any live mint or issuance path that can print supply is a loaded gun — hard caps, multi-sig, and renounced or tightly gated mint authority matter more than branding. Pulse lesson: treat owner/mint keys on tokens, bridges, and teleports as fatal if an EOA or low-threshold set can still call them.
9 Aug 2026: BlockWatchdog mapped ~$8.07M drained from Coinsbuy platform wallets — ~$6.04M USDT on Tron and ~$1.89M USDT + 77 ETH on Ethereum — linked into one operation via Bridgers. Most proceeds moved through FixedFloat; ChangeNOW froze a six-figure slice after investigators contacted them. Within about a day Coinsbuy refilled the drained wallets near prior balances, which researchers read as the team not treating this as a simple private-key wipe. Exact entry path (withdrawal system vs other) is still open. Confirmed on-chain forensics + major-press pickup — not a rumor thread.
369 take: Hot wallets and multi-chain withdrawal rails are still a primary loss surface even when smart contracts look fine. Rapid refill can protect users and still leave the entry path unexplained — watch for fake “Coinsbuy claim / migrate” phishing while recovery chatter is hot. Pulse lesson: minimise hot keys on ramps and bridges; anomaly response beats hoping keys were “probably fine.”
7–8 Aug 2026: SlowMist and Defimon/Clara flagged ~29,984 USDC drained from Atomic / AtomicLending on Arbitrum (Uniswap V3 LP + leveraged trading). A manager signature lacked binding to a specific position, caller, nonce, deadline, or chain — so the same approval could be reused across many position IDs. Combined with a same-transaction flash move of the ARB/USDC Uniswap V3 pool used for valuation, LP positions were burned for more value than they should have paid. Small dollar size; clean “unsigned context + spot price” lesson. Firm alerts cited — treat as research snapshot, not a full 369 deep-dive of the protocol.
369 take: Off-chain signatures for LP or position actions must name exactly what they authorize — position ID, nonce, deadline, and chain — or one stamp can be reused. Don’t price critical burns or payouts off a raw same-tx spot pool. Pulse lesson: any signed stake, ramp, or position manager inherits the same gun.
2 Aug 2026: Defimon flagged ~$690k (~693.5k USDC) drained from LpdFi / LOOPSDAO on BNB Chain in one transaction. The protocol priced LPD from live PancakeSwap LPD/USDC reserves with no time-weighted or manipulation guard. An attacker warped that pool in the same transaction, made interest accounting look far larger than the collateral, then was paid by burning the protocol’s own LP position and withdrawing USDC. Confirmed firm alert + BscScan success tx — textbook flash-loan + spot-oracle pattern, not a rumor thread.
369 take: If a protocol reads price from a live DEX pool and pays from shared LP or stables in the same window, flash loans can print fake value. Prefer time-weighted or multi-source prices, hard deviation bands, and claim caps — treat spot reserves as untrusted. Pulse lesson: same Allbridge-shaped gun on any PulseX / UniV2-style quote used for mint, interest, or redeem.
A third-party deposit rail for the AFX perp DEX was emptied of about 24.15 million USDC. Arbitrum’s own native bridge was not hit. Offchain Labs confirmed the bad transaction targeted AFX’s contract, not chain infrastructure.
Update (2026-08-03) — Goodwill / recovery plan window (Aug 3): AFX said a goodwill plan for users hit by the ~$24.15M custody-bridge drain would be unveiled around 3 Aug 2026. Team post-mortem line: attack started with social engineering of a developer, then supply-chain / internal tooling, then compromised validators co-signing the bridge withdrawal — not a break of Arbitrum’s native bridge. Treat compensation details as TBD until the official plan posts; fake “AFX claim / migrate” sites are phishing risk while recovery chatter is hot.
369 take: “On Arbitrum” often means a custom deposit bridge, not the canonical bridge. Short dispute windows and hot-validator-signed withdrawals are a full trust surface — treat them like mint authority before you deposit. Off-chain ops compromise (social eng → build/deploy → validators) is as deadly as a bad Solidity bug.
~1 Aug 2026: Onchain Lens (with SlowMist first-spot) reported 16.623 WETH (~$31k) drained on Base after an unverified helper spent a victim’s existing WETH allowance. Flow: victim → unverified spender → fresh helper → attacker wallet; funds still sitting with the attacker at report time. Victim publicly offered a white-hat split; no refund observed yet. Not a named blue-chip vault — residual approve on a contract with an open external-call surface. A near-identical attempt the next day only produced zero-value transfers. Confirmed firm alert + Basescan tx.
369 take: Old unlimited (or leftover) approvals on unverified routers and “call anything” helpers are still a live drain path — even at $30k on L2s. Pulse lesson: revoke stale allowances, never leave spend rights on sketchy helpers after you walk away from a UX, and treat unverified multicall-style contracts as hostile until proven otherwise.
~30 Jul 2026: SlowMist flagged a ~$9.6k drain on Index Coop / Set Protocol’s public ExchangeIssuance helper. An attacker used a malicious Set and manager hook so component amounts changed between quote and pull, then emptied residual inventory that had sat on the helper. Small dollar size; clean check-then-act pattern. Confirmed firm alert + explorer tx — not a rumor thread.
369 take: Public helpers that leave leftover tokens and trust live state between check and transfer are loaded guns even at tiny TVL. Pulse lesson for any issuance / swap / “issue for exact” router: lock what you quoted, don’t park dust, don’t let hooks rewrite amounts mid-flow.
Verified (29 Jul 2026): Academic paper ProxyMark (arXiv:2607.07062, submitted 8 Jul) shows a network-layer attack on Monero nodes that use Tor hidden-service peers. Originated txs are first sent to two Tor HS “proxy” peers before clearnet Dandelion++ — an adversary who occupies those outgoing peers can capture origin txs, then watermark traffic to link onion identity to a real IP. Authors report strong lab metrics (100% onion-ID precision; 7–11 of 12 outgoing HS links occupied in tests; watermark ~91–94% recall) on live Tor + Monero mainnet/testnet. This does not break ring signatures, stealth addresses, or RingCT, and does not read tx contents. Threat model needs a heavy Sybil/Tor-entry adversary — not “Tor is useless for everyone.” Monero Research Lab agenda for Wed 29 Jul 17:00 UTC explicitly lists this paper. Lineage: WWW’24 workshop abstract; full ProxyMark arXiv is the Jul 2026 writeup.
369 take: Privacy stacks fail at seams. Monero’s crypto can hold while Tor+P2P forwarding still leaks who originated a broadcast. Pulse lesson for any “private” path (mixers, shields, Tor wallets): ask what metadata bookends remain public and who can sit on the relay choke points.
Developing (28 Jul 2026 ~16:23 UTC): Blockaid reported an ongoing exploit on CryptoDAOGlobal’s Pro token on BNB Smart Chain and said ~$8.2M USDT sat with the exploiter plus three “winning” wallets at alert time. Token 0x8d657445…df0e2 (Pro Token); exploiter 0x427671b2…F45D; example tx hits Pancake Pro/USDT and Pro/CDAO pools. Our ~1h spot-check found only ~$1.0M USDT still on those four EOAs (BSC+ETH) — bags already moving, so treat $8.2M as the alert-time headline, not a final loss. Owner looks renounced; root cause not confirmed. No Pulse footprint. Do not ape into “recovery” DMs.
369 take: Renounced ERC20 + audit sticker ≠ safe if pool/prize/related mechanics can still yank USDT. Pulse lesson: high-hype “DAO/Pro” branding does not replace reading where liquidity actually sits. Not a Pulse deep — cross-chain awareness only.
23 Jul 2026 (verified chase 28 Jul): Solido Cash on Supra (Move) was hit via an oracle misassignment that valued collateral near ~$1 while market was a fraction of that. Attacker minted CASH against the bad price, sold for SUPRA. Solido forensic (via AMBCrypto 27 Jul): two waves — one atomic, then five wallets — minted ~809k CASH and ~293.7M SUPRA net proceeds; ~84% traced to CEX infrastructure; containment disabled the mint path. DeFiLlama hacks table lists ~$73.4k USD (Protocol Logic / Oracle Misconfiguration). Grok daily’s ~$900k figure is overstated vs Llama — prefer forensic token totals + Llama USD. Distinct from Bonzo Lend / Supra Hedera ~$9M (11 Jul) zero-signature verifier bug.
369 take: Oracle config / assignment errors print money the same way bad feeds do. Pulse CDP/lending lesson: wrong collateral mark + weak mint caps = loaded gun. Do not merge with the separate Supra-on-Hedera verifier bug that hit Bonzo.
Developing (26–27 Jul 2026): Blockaid flagged an exploit on @gardenfi HTLC rails — ~$450k USDT across Ethereum, Base, Arbitrum, and BSC. Exploiter EOA 0x25b224c05f6cc5e132165c1621de1a4c3b316999 (~$425k / ~20 txs at alert). Update (26 Jul 18:10 UTC): Garden posted that they identified unusual activity, took the app offline for a full investigation, and will share more when they have it. App shows “Garden Maintenance / under maintenance.” No official loss figure or root-cause post yet; blog has no incident article. Not the Oct 2025 ~$11M+ solver compromise — do not merge the two. Updated (28 Jul 2026): Garden told Cointelegraph (story revised 27 Jul 02:16 UTC) that protocol + HTLC contracts were not compromised; an independent solver’s off-chain database was breached and fraudulent swap records inserted, causing that solver to release funds for unbacked swaps. Team line: solver-owned assets only, no user funds at risk; still confirming totals; working with zeroShadow, Quantstamp, Blockaid. Keep Oct 2025 ~$11M+ solver-environment hit as a separate incident.
369 take: Original Blockaid HTLC drain alert still stands as the on-chain symptom. Garden’s later line shifts root cause to an independent solver’s off-chain DB — still a fund-release path you have to trust. Pulse/ramp lesson: treat every off-chain solver/keeper/DB that can trigger release as adversarial, even when “core contracts are fine.” Not the Oct 2025 hit.
26 Jul 2026: WEMIX said ownership of a contract linked to its WEMIX$ stablecoin path was compromised. Attacker unauthorized-issued ~5.23M WEMIX$, swapped into ~30.7k WEMIX + ~724k USDC.e, then bridged proceeds toward Ethereum and BNB. Team suspended WEMIX3.0 bridges (incl. CCIP / PLAY), affected LPs, and related modules. Preliminary figures may change; bridge itself was not framed as the entry — the linked / ownership surface was.
369 take: Classic peripheral gun: a non-core Ownable that can still mint or authorize issuance. Pulse lesson for stables / migrations — inventory every linked contract with mint or owner powers, not just the “main” bridge. Not a Pulse deep; pattern card only.
We reviewed a 24 Jul X amplification (@CryptocapoOO) claiming Hyperliquid was being “hacked” via Google: search the brand → top Sponsored result → clone site → connect wallet → funds gone, “hundreds” emptied. Our check: the Google Ads → pixel-clone → approve/permit drain pattern is real and well documented for Hyperliquid (and Uniswap/Aave/Jupiter-class brands) across 2025–2026, with brand-impersonation domains still flagged in Jul 2026. Framing it as an Hyperliquid L1/app exploit is wrong — users are phished; contracts are not the breach. “Hundreds emptied this week” stays unverified without a firm address table. Official app to bookmark only: app.hyperliquid.xyz — never open from Sponsored results.
369 take: Sponsored search is not the official site. Same gun hits Pulse dApps and wallets — bookmark the canonical URL, ignore ads, and read every approval. If you already connected a fake front end, revoke allowances and move funds from a clean bookmark only. This is phishing, not “the protocol got hacked.”
Specter flagged multi-chain hot-wallet drains linked to @TripleH / @TripleAHQ (Triple-A payment rails). He reported >$9.3M drained/swapped/bridged to Ethereum. We independently confirmed the listed consolidation address 0x01F83B5d…253b1 holds ~5,228 ETH (~$9.72M) with no outbound sends — matching the ~5,227 ETH headline. Other listed ETH addrs are empty now. TRON/Solana addresses are listed but not re-balanced here.
Update (26–27 Jul): Desk reporting (The Block via Specter) revised the multi-chain sweep toward ~$11.8M as new deposits kept landing in compromised wallets; some trackers say ~$12M. We still do not invent a 369 dollar total beyond investigator ranges. Team said they are investigating and that customer funds are not impacted — formal map still thin. Consolidation ETH bag thesis for the listed EOA unchanged pending a full address table.
369 take: Payment-gateway hot wallets are custodial: you send to their address. If those keys move, many checkouts die together. Self-custody seedless wallets are a different product. Until the team freezes deposits and publishes a full map, treat Triple-A / TripleH deposit addresses as high risk.
Kraken Security Labs flagged a new underground offer: WARDEN (actor WardenStealer) — a Windows x64 stealer, clipper, and loader sold with a control panel (~$349/mo). Seller claims include Chromium/Gecko credential theft, 200+ cryptocurrency browser extensions, clipboard replacement of BTC/ETH addresses, App-Bound Encryption bypass, and secondary payload delivery. Kraken notes these are threat-actor marketing claims and have not been independently verified.
369 take: Extension wallets are the advertised target. Treat unknown Windows installs and “free crack” tools as hostile. Seedless / hardware-unlock wallets raise the bar against seed dump, but clippers still hit any pasted receive address — verify last characters before you send. Never download “samples” from underground ads.
SlowMist reported ~542k USDC lost after attackers abused Lien’s bond exchange path. We verified live BondMakerCollateralizedEth contracts on Ethereum expose exchangeEquivalentBonds(exceptionBonds[]) and permissionless registerNewBondGroup — matching both the multiset-validation and open-registration narratives. Victim address still holds residual USDC after the event. Same BondMaker family as the 2020 white-hat save; this time funds left.
369 take: Open bond/ticket registration plus soft equivalence or OTC pricing is a loaded gun. Before you LP into structured “bond” pools, ask whether anyone can mint paper the pool will buy without hard collateral checks.
Kaspersky researchers detailed OkoBot, a modular malware platform. Its SeedHunter module detects popular hardware-wallet software and shows fake recovery screens to harvest seed phrases. Delivery often uses social-engineering “fix this error” prompts and trojanized installers.
369 take: Never type a seed phrase into any on-screen prompt — not even one that looks like Ledger or Trezor. Hardware recovery stays offline. Seedless wallets remove that Notes/Keychain target, but malware can still abuse unlocked sessions — keep devices clean and ignore “run this command” fixes.