TL;DR
DeFi matured — and the threat moved. Losses fell 80% from $2.62B (2022) to $534M (2024); bridges are now just 3% of losses (was 73%). But 89.1% of 2025 losses are protocol-logic bugs — bespoke, hard to pattern-defend. Malicious approvals alone caused $1.46B. An audit is not a guarantee. A bridge is not a bank. An approval is not a formality.
| RISK | WHAT IT IS | 10-MINUTE VET |
|---|---|---|
| Malicious approval | Unlimited spend permission → instant drain | Check approvals. Cap + revoke. Simulate. |
| Bridge trust | Validators / multisig / logic can be compromised ($2.9B total) | Who validates? What size? What’s the pause? |
| Logic bug | Bespoke math/access flaw, multi-chain spread | Audits? Time? TVL trend? Incident history? |
| Oracle risk | Price feed manipulated → mispricing drain | Which oracle? Fallbacks? Deviation guards? |
| Rug / honeypot | Liquidity pulled or sells blocked | Liquidity locked? Owner renounced? Check on explorer. |
| MEV / slippage | Sandwiched or fat-fingered execution | Slippage caps. Private routing. Small test first. |
The bridge era produced the canonical failures: Ronin ($624M, 5-of-9 keys phished), Wormhole ($326M), BNB Bridge ($570M), Nomad ($190M). $2.9B across 34 incidents (2021–Q3 2025), with $1.84B from contract exploits and 87% involving mixers. Median loss per incident fell from $6M to $1.5M — scale is harder, but count rose to 83 incidents in 2025. More small cuts, fewer cataclysms. That is what maturing security looks like.
The 2025 lesson is shared-code risk: Balancer’s rounding flaw did not just hit one chain — it hit every chain with the same pools. A protocol deployed on six networks with shared logic gives one bug six blast radii. Vet the deployment graph, not just the audit PDF. Highly audited, battle-tested code still breaks (Balancer, Yearn). Audits narrow the window; they do not close it.
Approvals are the retail-scale version of the same failure. A single “max approve” to a drainer is a full wallet loss with no exploit needed — the user authorized it. Default to exact-amount approvals, revoke after use, and simulate any new contract interaction. If a frontend demands unlimited approval to “enable trading,” that is a design smell. Leave.
2022 PEAK
Bridge era
$2.62B losses. Bridges 73%. Ronin, Wormhole, Nomad dominate the ledger.
2024 MATURE
Engineering wins
$534M losses. Bridges 3%, flash-loans <1%. Median incident $1.5M. Repeatable bugs engineered out.
2025 NOW
Logic era
$680M, driven by few large logic bugs. 89.1% protocol-logic. Multi-chain spread is the multiplier.
THE 10-MINUTE PROTOCOL VET
1. Contract verified + audited (by whom, when). 2. Admin is multisig with timelock. 3. TVL trend is real, not incentivized. 4. Bridge/validator model is named. 5. You can simulate, pause, and revoke. Fail any one → size down or skip. Yield is not a reason to skip the vet.
DEFENSE CHECKLIST — MODULE 06
[01] Cap approvals. Revoke unlimited. Simulate before signing any new contract.
[02] Vet every protocol in 10 minutes: audit, multisig, validator model, pause path.
[03] Size bridges by trust, not yield. Assume validator compromise is possible.
[04] Cap slippage, use private routing for size, test with small amounts first.
[05] Walker the deployment graph: shared logic × N chains = N× blast radius.
SCHOOL INDEX
← PREVIOUS · MODULE 05
Privacy Is Not Pseudonymity
How chain analytics clusters your wallets, why Tornado Cash still leaks 34.7%, and what real OPSEC looks like on transparent chains.
NEXT · MODULE 07 →
The Human Layer
Pig butchering, romance-investment fraud, crypto ATMs, and why 78% of victims never knew — plus verification rituals that actually work.
PNTHR SCHOOL · FREE + OPEN · SOURCES: FBI IC3 2025 · CHAINALYSIS 2026 · TRM LABS · CERTIK · IMMUNEFI · TUTELA · EDUCATION ONLY · NOT FINANCIAL ADVICE · NOT LEGAL ADVICE