Three unrelated bugs have drained DeFi protocols since October 1, 2026, and none of them touched a headline-grabbing exchange. A MoveVM flaw let an attacker mint unbacked balances on Initia, which halted its Layer 1 and rollups on October 23. A share-accounting failure in UNCX Network’s Smart Token Vesting V2 spread across four chains. A spoofable caller check in FlashLoopAdapter, a third-party Safe module, cost two multisig wallets about 114 ETH.

Reported losses run from roughly 114 ETH to about $2.7 million, small next to the nine-figure thefts of past years. The pattern matters more than the totals. Each incident sat at a different layer of the stack (virtual machine, application contract, wallet module), and each slipped past the checks teams usually trust. This is the state of these DeFi exploits as of November 11, 2026, including what is confirmed, what is still disputed, and what builders and treasury managers should do about it. Initia’s trouble also follows the cluster of bridge hacks involving Cysic, WAX and Initia reported in late October.

Don't miss new tech stories on Google

Add Shattered once in the Google app and our stories show up in your news suggestions.

Timeline of the Three DeFi Exploits and the Responses

The sequence runs from October 1 to November 10, with defensive product launches landing in the middle of it. Dates below come from incident trackers, protocol notices and vendor announcements. Where a later report revised an earlier figure, the table keeps the more conservative number.

Read top to bottom, the table shows how little time defenders got. Safe shipped Safe Pro on October 6 and Safenet on October 21 while FlashLoopAdapter, UNCX and Initia were all unresolved. None of those products was designed in response to these three bugs, but each addresses the same family of failures: unauthorized calls, unchecked transaction contents and weak signing context.

Initia: How a MoveVM Bug Minted Unbacked Balances

Initia is a Layer 1 with rollups, built on the Cosmos SDK with a MoveVM execution layer. On October 23, it stopped the L1 and every rollup after reports that a vulnerability in the MoveVM implementation allowed an attacker to create unbacked balances for WETH, AUSD0 (also referred to as iUSD) and USDC. Available reporting estimates that about $2.7 million in assets moved out before the halt. One outlet put cash-outs nearer $2.6 million, and no final figure has been published.

The mechanics matter for every chain that bridges assets. Wrapped tokens on a rollup are only as good as the backing held on the other side. If a VM bug lets an attacker print balances, the printed tokens can be redeemed through a bridge against real reserves, which is what drained value here. Reporting says Initia identified the vulnerable MoveVM call logic, prepared a targeted patch and sent it for external security review instead of rushing a hard fork.

The Oct. 30 Recovery Plan and the Nov. 2 Restart

On October 30, Initia announced that the network would resume on November 2 after applying an independently audited MoveVM patch. The plan isolated the remaining forged balances, restored backing for AUSD0, USDC.n and WETH, and kept transaction history intact with no rollback. Staged validator upgrades, post-restart monitoring and DEX pool rebalancing were part of the same package. Coverage of the restart says LayerZero withdrawals of WETH and AUSD0 carried a 24-hour delay while Noble USDC withdrawals faced an hourly limit.

Those throttles trade user convenience for time. A 24-hour delay gives operators a window to spot a bad withdrawal before it settles, a control worth studying in any bridge verifier settings audit. TokenPost reported on November 2 that the L1 and rollups returned to normal operation with transfers facing limits.

The Nov. 10 Halt Points to a Different Attack Path

Eight days after the restart, Initia halted the L1 again. Initia said compromised rollup infrastructure keys were used to forge IBC transfers, according to reports, and about $605,000 in USDC plus $10,000 in other tokens left a custody account. The company said it had found no evidence that user funds were affected so far. A separate disclosure circulating on November 9 and 10 cited a figure near $3.27 million. Because the two numbers conflict and no post-mortem exists yet, the conservative reading is that the second incident involved at least $615,000 and that the final total is unsettled.

The distinction is important. The October bug lived in code, and an audited patch addresses it. The November incident, as described, involved keys, and a patch cannot fix a key that an attacker already holds. Treating the two as one story would hide that Initia now faces both a software-assurance problem and an operational-security problem.

UNCX Vesting V2: A Share-Accounting Bug Across Four Chains

UNCX Network, formerly known as Unicrypt, runs token-locking and vesting products used by project teams to prove that team and investor tokens cannot be dumped early. Its Smart Token Vesting V2 contracts were exploited on October 14. The root cause was inconsistent share accounting: when vesting positions were released, topped up and canceled in certain sequences, the contract’s internal records stopped matching the token balance, and the attacker could withdraw more than the position entitled them to.

The early picture looked small. A first Ethereum drain reached roughly $86,000. As the same flaw was found on other deployments, BTCUSA’s tally of the four-chain drain climbed to about $1.49 million across Ethereum, Base, BNB Smart Chain and Robinhood Chain. UNCX published an official cross-chain incident report on October 15 and updated it on October 17. A standard smart contract audit exists to catch exactly this kind of state-desync bug, so its survival into production is the uncomfortable part.

Treat the $1.49 million as an early external estimate dated October 15 and 16. Later reporting says UNCX’s own figure for realized extraction was lower, around $807,000 to $808,000, with about $58,000 recovered by whitehats. Those lower numbers come from a single report, so they carry a hedge. The spread between $1.49 million and roughly $808,000 is a reminder that gross drains, net extraction and recoveries are different measurements and get mixed up in headlines.

As containment, UNCX reportedly raised the fixed fee for new locks to more than 1,000 ETH, removed fee-bypass paths and pushed token fees to the contract maximum of 5%, which in practice makes Vesting V2 unusable for new customers. A report dated November 9 says UNCX was coordinating with law enforcement and the SEAL 911 emergency group, and that neither the original attacker nor copycats had returned funds. The flaw is unrelated to the FlashLoopAdapter bug, despite both surfacing in the same month.

FlashLoopAdapter: A Safe Module That Trusted the Caller

FlashLoopAdapter is a third-party Safe module that lets users open and close leveraged Aave V3 positions using flash loans. Neither Aave nor Safe maintains it. A security analysis dates its deployment to July 2026. On October 1, an attacker deployed a fake Safe contract that answered “yes” when the module asked whether it was enabled, then used that false answer to make two real Safes execute unauthorized transactions. QuiverCrypto’s write-up puts the loss at about 114 ETH.

The flaw is a classic confused-deputy design. The module called ISafe(msg.sender).isModuleEnabled(address(this)) and believed the reply, without checking that the caller was a real, authorized Safe. It also accepted a caller-supplied router and calldata with no effective allowlist. Anyone could therefore present a contract that lies about its status and point the module at an arbitrary call. This was an access-control failure, and the access control testing approach that catches it is cheap compared with the damage.

The Vulnerable Pattern in Simplified Form

Public technical analyses recommended exactly this shape: independently register authorized Safes and restrict router destinations or calldata. The same module-authorization theme runs through a recent walkthrough of smart account module exploits in Foundry. A historical summary says the affected feature was removed on October 7 after a second critical bug report, but that date and its scope remain unofficially reported. As of November 9 to 11, incident trackers still list FlashLoopAdapter as hacked, and no patched redeployment, rename or successor has been confirmed.

Why Aave V3 Was Not the Target

Headlines that label the October 1 incident an “Aave exploit” mislead readers. The reporting describes FlashLoopAdapter as a third-party tool that happened to operate on Aave V3 positions. Aave V3’s core contracts were reported unaffected, and Aave founder Stani Kulechov was reported as saying the incident did not involve those core contracts. No Aave loss, patch or emergency response tied to this incident has been confirmed.

Aave’s own security calendar looks routine by comparison. Version 3.7 completed its rollout in October, and an ARFC for version 3.8 was published on November 2, described as a non-invasive, security-focused release. V3.8 is proposed and heading into audits by Certora and others before any governance vote, so it is not yet released. On November 6, Aave Labs announced that it had acquired the team and technology behind Avon, a fixed-rate onchain lending project. That deal does not replace or rename V3. Aave also reported that an AI-assisted review of V3 and V4 produced 71 findings with no Critical or High-severity issue confirmed, though that figure rests on a single report.

The lesson for integrators is blunt. A blue-chip protocol’s audit history does not transfer to the add-ons built on top of it. Users who connect a Safe to an unaudited module inherit that module’s risk, whatever the lending market underneath looks like.

Safe’s Response: Safenet, Safe Pro and the Policy Engine

Safe did not confirm that Safe itself was exploited in any of the three incidents, and available official results do not suggest it was. The company still used October to push new defenses. Safe Pro launched on October 6 as a paid organizational layer on top of the still-free Safe{Wallet}. Starter costs €189 per month or €1,990 per year, and Business starts at €669 per month or €7,190 per year, all excluding VAT. Enterprise pricing is custom.

Safenet is the more relevant launch. Announced on October 21, it lives in Safe{Shield} and submits a transaction to a jury of independent security firms (BlockSec, Candide, IntentGuard and OpenCover) before it executes, then records the jury’s verdict onchain. It runs for Safe Pro users on Ethereum, Arbitrum and Gnosis Chain, with broader visibility for Safe multisig users planned for 2027. In its launch announcement, Safe Labs said accounts on its platform have processed more than $1 trillion in transactions and that attacks aimed at them are getting cheaper to run.

A Safe Policy Engine was announced on October 27 for Q4 2026, according to a single source, and would let teams enforce transaction rules through smart contracts, with a co-signer for rules a blockchain cannot evaluate. It targets Business and Enterprise customers, and there is no confirmation yet that it has shipped. Pre-execution screening would not have fixed FlashLoopAdapter’s code, but it targets the failure mode behind the October 1 drain, in which a transaction looked legitimate to the module and malicious in effect.

What the Numbers Do and Do Not Prove

Three numbers dominate coverage, and each needs a label. The $2.7 million Initia figure is an estimate from preliminary reports, and the sourcing does not establish it as the final loss. The $1.49 million UNCX figure is an early external estimate that UNCX’s later reporting appears to undercut. The 114 ETH FlashLoopAdapter loss converts to roughly $300,000 to $310,000 depending on valuation date, and a fiat value pinned to one day should not be read as final.

Two further claims deserve caution. No source confirms that the Initia exploit directly affected the INIT token’s supply, so any statement along those lines is unproven. And no credible link exists among the three incidents, because they involve different teams, code bases and attack techniques. Treating them as one campaign inflates the story and misleads defenders about what to fix.

What the numbers do show is a cost curve. Each of these DeFi exploits cost an attacker little more than reading the code. Smart contract vulnerabilities in verified, publicly readable contracts get found by anyone patient enough to look, and the defenders’ disadvantage grows with each new module, adapter and bridge a stack adds.

Market Impact: Price Spikes, Fee Shocks and Frozen Withdrawals

The market reaction to Initia was counterintuitive. After the November 10 halt, outlets reported INIT up sharply over 24 hours, with gains cited between roughly 60% and 80% depending on the outlet and time window. Headlines like that reward thin liquidity more than fundamentals, and a halted chain cannot clear normal arbitrage, which tends to produce disorderly prices rather than informative ones. Anyone reading the move as a vote of confidence should note the chain was halted at the time.

The harder cost is operational. Initia users face withdrawal delays and hourly caps that persist after a restart. UNCX customers face lock fees so high that the product is effectively closed to new use. Projects that locked team tokens in Vesting V2 now carry a trust problem they did not create, since their investors can no longer assume the lock works as advertised. These frictions rarely appear in loss tables but they shape where capital and developers go next.

For a broader view of the crypto sector’s security backdrop, the cryptocurrency section tracks the larger incidents of 2026, which dwarf these three in dollars even if not in lessons.

Competitive Comparison: Three Bug Classes, Three Different Fixes

Comparing the incidents side by side shows why no single control would have stopped all of them. The table uses the most conservative reported figures.

The fixes diverge. Initia needed a VM-level patch, an external audit and key hygiene. UNCX needed invariant checks that tie internal share totals to token balances. FlashLoopAdapter needed an authorization registry. A team that only buys one scanner or one audit misses at least two of the three classes.

Historical Context: Bridges and Modules Keep Failing in 2026

These incidents did not arrive in a quiet year. Ankr’s Ethereum bridge lost over $303,000 on November 9 after five suspicious withdrawals linked to a previously redeemed Optimism deposit, a case covered in detail in the report on the Ankr bridge forged receipts. Earlier in the autumn, a bridge security study of 36 protocols flagged widespread weaknesses in how cross-chain message validation is built and tested.

The common thread across 2026 is trust placed in the wrong object. Bridges trusted receipts they could not prove. The FlashLoopAdapter module trusted a caller’s self-report. UNCX trusted its own bookkeeping. Initia’s VM trusted a call path that a mint could slip through. Each case replaced independent verification with an assumption that held until someone tested it.

That is also why post-mortems matter. As of November 11, no result confirms that Initia has published a final post-mortem, a complete reimbursement record or a firm statement on the November incident. Until one appears, the community works from fragments.

What Developers and Treasuries Should Check This Week

Teams do not need new tools to act on these cases. They need to apply tests that already exist to the layers they usually skip.

  • Inventory every Safe module and adapter enabled on treasury multisigs, and disable any unaudited or unmaintained one. Third-party code with spend authority is the highest-risk item on the list.
  • Write invariant tests that tie internal accounting totals to actual token balances, and run them against release, top-up and cancel sequences. A DeFi exploit monitor can alert on balance divergence in production.
  • Verify that any contract calling isModuleEnabled, or any similar self-reported status, checks the caller against a registry the contract controls.
  • Restrict router and calldata targets for any function that makes arbitrary external calls, and use cross-chain bridge exploit testing in Foundry to confirm forged messages are rejected.
  • Rotate and segregate rollup infrastructure keys, and place delay or rate limits on custody accounts that hold backing assets.
  • Review the bridge security testing steps against any bridged asset your protocol accepts as collateral.

Predictions: Five Things to Watch Through Early 2027

These are editorial judgments, not reported facts.

  • Initia publishes a full post-mortem. The conflicting loss figures and the shift from a code bug to compromised keys create pressure for a single authoritative account. Expect it before any serious return of liquidity.
  • Module allowlists become a default. After FlashLoopAdapter, wallet interfaces and DAO tooling will likely flag unaudited Safe modules more aggressively, and some treasuries will ban them by policy.
  • Vesting and locking products face re-audits. UNCX’s problem casts doubt on how every locker handles release, top-up and cancel logic, so competing platforms will likely commission fresh reviews to reassure customers.
  • Pre-execution screening spreads. If Safenet and the Policy Engine reach general availability on schedule, other smart-account vendors will copy the jury and policy model to stay competitive.
  • Loss tables get more careful. Analysts and aggregators will need to separate gross drains, net extraction and recoveries, as the UNCX gap showed, or their monthly totals will keep drifting from reality.

The Takeaway

Three modest-sized DeFi exploits exposed three distinct failures: code in a virtual machine, bookkeeping in an application contract and trust in a wallet module. Total reported losses stay in the low millions, but the repeated pattern of unverified assumptions is the real signal. Teams that test accounting invariants, restrict module authority and treat keys as a first-class attack surface will be harder targets than those waiting for the next audit cycle.

FAQ

What caused the Initia exploit?

Reports describe a MoveVM vulnerability on a Cosmos SDK-based chain that let an attacker create unbacked balances for WETH, AUSD0 and USDC. Initia halted its L1 and rollups on October 23 and later planned a restart with an independently audited patch.

How much was stolen in each incident?

Reported figures are about $2.7 million for Initia’s first incident, about $1.49 million as an early estimate for UNCX Vesting V2, and about 114 ETH for FlashLoopAdapter. All three are estimates, and UNCX’s later figure for realized extraction was lower, according to a single report.

Was Aave hacked?

No. The FlashLoopAdapter incident involved a third-party Safe module used with Aave V3 positions. Reporting says Aave V3’s core contracts were unaffected, and no Aave loss or emergency patch has been confirmed.

Was Safe compromised?

Official Safe results do not confirm that Safe itself was exploited. Two Safe multisig wallets were drained because they had enabled a vulnerable third-party module, not because of a flaw in Safe’s own contracts.

Did Initia get hit again in November?

Reports say the L1 halted again on November 10 after unauthorized transfers from a custody account, with compromised rollup infrastructure keys used to forge IBC transfers. About $615,000 is the lower reported figure, while another disclosure cited roughly $3.27 million, so the total is unsettled.

Are the three incidents connected?

No evidence links them. They involve different teams, code bases and root causes, and UNCX’s accounting flaw is a separate vulnerability from the FlashLoopAdapter authentication bug.

Is UNCX Vesting V2 safe to use now?

It appears operationally contained rather than fixed. UNCX reportedly raised new-lock fees above 1,000 ETH and set token fees at the 5% maximum, which blocks normal use, and no relaunch has been confirmed.

Ryan Cole writes about security engineering and protocols: how HTTPS, certificates, and modern encryption protect data in transit, explained without the jargon.