RWA Hacks and Exploits: Losses and Root Causes Since 2023
Quick answer. By our count of DefiLlama's data, real-world asset protocols lost $11.73 million in seven incidents since 2023, a small sample. A Zoth admin key compromise in March 2025 caused 75%; five incidents were contract bugs, one address poisoning.
Video transcript
Real-world asset protocols lost 11.73 million dollars in seven incidents since 2023, by our count of DefiLlama's data. One admin key compromise caused three quarters of it.
Seven is a small sample, and hacks data leaves out credit and legal failures. Low losses are not proof of low risk.
2025 holds 78 percent of the total, almost all from Zoth. Of all hack losses since 2023, RWA protocols are 0.08 percent.
Zoth is the case that matters. An attacker used compromised admin keys to upgrade a vault proxy. DefiLlama records 8.85 million dollars, Rekt about 8.4 million.
Contract bugs are five of the seven incidents, but one key compromise is most of the money. No incident carries an oracle label.
Every contract bug sat in code around the asset: a collateral check, a vault, credit pools, a proxy and a yield feature. No record shows the asset itself failing.
The controls are specific. Put upgrades behind a multisig and timelock, price collateral from a manipulation-resistant oracle, initialize every proxy, pause and empty deprecated contracts, and pay only allowlisted addresses.
Guardian's Vanguard audit pairs expert manual review with Helix, our AI auditor. After launch, Managed Security monitors every block and reviews every multisig transaction before signing.
Read the full article for every incident and its control, or talk to Guardian about closing these gaps.
Tokenized assets are moving onto public chains, and risk teams want to know how they fail.
So we counted RWA incidents in DefiLlama's public hacks dataset, read on November 6, 2026. Every total here is our computation from that data, which anyone can rebuild.
Seven incidents is a small sample. Read the totals as a record of what happened, not as a measure of how safe RWA protocols are.
How We Selected RWA Incidents
DefiLlama marks real-world assets in two places: a hack's target type, and the category of the protocol a hack is linked to.
We counted an incident if it met one of three tests:
- Target type "RWA": one record, Aurellion.
- Linked protocol in the RWA category: Florence Finance, Bloom, Usual USD0 and Huma.
- RWA product family: the linked protocol shares a parent with an RWA product, and DefiLlama describes it as RWA. This adds Zoth's two incidents (ZeUSD, filed under CDP).
Resolv USR and SolvBTC also share a parent with an RWA product. DefiLlama does not describe either as RWA, so both are out.
Data and limits
- Data: DefiLlama's hacks and protocols endpoints, read November 6, 2026. The latest hack entry is November 5.
- Amounts are DefiLlama's reported losses in US dollars, before recoveries.
- Root causes follow DefiLlama's labels: Key Compromise is keys, Social Engineering is offchain, code labels are contract bugs.
- Not counted: CrediX ($4.5 million, August 2025, Sonic) has no protocol link. The Credix in DefiLlama's RWA list is a Solana platform.
- Also outside the tests: Goldfinch ($330,000, filed as Uncollateralized Lending) and Swarm Markets ($472,892, filed as Dexs).
- Sample size: seven incidents. Percentages describe these seven, not a trend.
One link is doubtful. DefiLlama ties the Bloom incident, on Blast, to an RWA protocol it lists on Ethereum only. It also lists Bloom Trading, a Blast trading platform, which may be the real victim.
We kept DefiLlama's link, as our tests require. Without Bloom, the totals are six incidents and $11.19 million.
RWA Losses by Year
| Year | Incidents | Reported losses | Largest |
|---|---|---|---|
| 2023 | 1 | $1.45M | Florence Finance |
| 2024 | 1 | $0.54M | Bloom |
| 2025 | 3 | $9.18M | Zoth ($8.85M) |
| 2026 to Nov 5 | 2 | $0.56M | Aurellion |
| Total | 7 | $11.73M | Zoth |
2025 holds 78% of the total, almost all of it from Zoth's March key compromise.
For scale, DefiLlama records $14.79 billion across 894 hacks from January 2023 to November 5, 2026. RWA protocols are 0.78% of those incidents and 0.08% of the value.
Low losses are not proof of low risk. The sample is small, and a hacks dataset does not count credit defaults or legal failures.
Every Incident and Its Root Cause
| Date | Protocol | Loss | Root cause | What happened |
|---|---|---|---|---|
| Nov 30, 2023 | Florence Finance | $1.45M | Offchain | USDC sent to a lookalike address |
| May 9, 2024 | Bloom | $540K | Contract bug | Yield reset flaw on Blast (ChainCatcher); 90% recovered |
| Mar 1, 2025 | Zoth | $285K | Contract bug | Collateral valued after a flash-loan-skewed swap |
| Mar 21, 2025 | Zoth | $8.85M | Keys | Admin access used to upgrade a vault proxy |
| May 27, 2025 | Usual USD0++ vault | $43K | Contract bug | Missing validation on capped unwrapping |
| May 11, 2026 | Huma | $101K | Contract bug | Open functions in deprecated V1 credit pools |
| May 12, 2026 | Aurellion | $456K | Contract bug | Uninitialized proxy let the attacker take ownership |
Zoth is the case that matters. Rekt reports the attacker used compromised admin keys to upgrade the USD0PPSubVaultUpgradeable proxy, then withdrew 8.85 million USD0++.
DefiLlama records that as $8.85 million; Rekt valued it at about $8.4 million.
On March 1, Zoth had lost $285,000 to a different flaw. Per DefiLlama, its loan-to-value check trusted a Uniswap V3 swap that a flash loan had skewed.
Florence Finance's loss was offchain. PeckShield reported on November 30, 2023 that about $1.45 million in USDC went to a phishing address instead of the intended one.
Bloom said it recovered $486,000, 90% of the $540,000 taken, after paying a 10% bounty, according to ChainCatcher.
Keys, Code and Oracles
| Root cause | Incidents | Losses | Share of value |
|---|---|---|---|
| Keys | 1 | $8.85M | 75.5% |
| Offchain (phishing) | 1 | $1.45M | 12.4% |
| Contract bug | 5 | $1.43M | 12.2% |
| Oracle | 0 | $0 | 0% |
| Legal or credit failure | Not tracked | n/a | n/a |
Code bugs are most of the incidents; one key compromise is most of the money. With seven incidents, a single large loss decides the split.
In our ranking of the biggest crypto hacks, most of the money also left through keys and people, not code.
No incident carries an oracle label. Zoth's March 1 loss came closest: it trusted a manipulable swap price, which DefiLlama files as protocol logic.
Per the incident descriptions, all five contract bugs sat in code around the asset: a collateral check, a vault, credit pools, a proxy and a yield feature.
Two hit code that was easy to overlook: Huma's deprecated V1 pools, still holding residual balances, and Aurellion's partly initialized proxy.
What the tokenization layer got right
None of the seven records describes a failure of the underlying asset, its custodian or its register. The losses came from contracts and wallets around the asset.
Issuer controls also limited damage. Rekt reports that asset issuers locked down 73% of Zoth's TVL right after the breach.
Per DefiLlama's description, no user funds were affected in the Usual vault incident.
Controls That Address Each Incident
| Incident | Control |
|---|---|
| Zoth, admin key | Upgrades behind a multisig and timelock, with alerts on every upgrade |
| Zoth, collateral value | Price collateral from a manipulation-resistant oracle, never a same-block swap |
| Aurellion | Initialize every proxy and facet at deployment, then verify the owner |
| Huma | Pause and empty deprecated contracts; restrict every state change |
| Usual | Validate caps and inputs on every deposit and unwrap path |
| Bloom | Fuzz yield reset and accounting invariants under extreme conditions |
| Florence Finance | Pay only allowlisted addresses; never copy from transaction history |
Our guides cover two of these in depth: initializer vulnerabilities and address poisoning.
Buying or building on a platform? Our tokenization platform due diligence checklist turns these into questions. Our RWA tokenization guide maps the full stack.
How Guardian Helps
Guardian's Vanguard audit runs two teams on the same code: expert manual review with invariant fuzzing, and Helix, Guardian's AI auditor.
After launch, Managed Security monitors contracts on every block, including upgrades, role changes and large fund movements.
Its security operations team reviews every multisig transaction before it is signed.
Frequently Asked Questions
How much has been lost to RWA protocol hacks?
DefiLlama's hacks data, read November 6, 2026, records seven incidents at real-world asset protocols since November 2023, totaling $11.73 million. That is 0.08% of all hack losses DefiLlama records from 2023 on.
What was the biggest RWA hack?
Zoth, on March 21, 2025. An attacker with admin access upgraded a vault proxy and withdrew 8.85 million USD0++. DefiLlama records the loss at $8.85 million; Rekt valued it at about $8.4 million.
Are RWA protocols hacked more through code bugs or keys?
By count, code bugs: five of the seven incidents. By value, keys: one admin key compromise, at Zoth, is 75% of the total. Seven incidents is too few to call this a trend.
Do these figures include credit defaults or legal failures?
No. DefiLlama's hacks dataset covers exploits and thefts. Borrower defaults, originator fraud and legal disputes are separate RWA risks and are not in these figures.
Sources
- DefiLlama: Hacks dataset API (read November 6, 2026)
- DefiLlama: Protocols API, including the RWA category (read November 6, 2026)
- Rekt: Zoth, Rekt (March 24, 2025)
- Rekt: Credix, Rekt (August 5, 2025)
- CryptoPotato: RWA Protocol Florence Finance Loses $1.45M in Address Poisoning Attack (November 2023)
- ChainCatcher: Bloom: 90% of the stolen funds have been recovered (May 10, 2024)
- Guardian: The Biggest Crypto Hacks of All Time, Ranked
- Guardian: Managed Security
- Guardian: Vanguard
Close the gaps behind RWA losses
Guardian audits the vaults, proxies and pools around your asset, then watches its keys and upgrades after launch.