Taiko Bridge
Why it happened
Attack class: Access controlA leaked enclave signing key and a registration check that allowed insecure enclaves let rogue provers finalize a fake Taiko state on Ethereum, which the bridge honored.
What would have stopped it
- Keep signing keys out of source control and scan repositories for secrets.
- Reject debug-mode enclaves at registration.
Not scored before the exploit
BlackHart had not scored Taiko Bridge before this exploit.
An attacker took about $1.75 million from Taiko's bridge on Ethereum by getting rogue provers accepted and using them to finalize a fake version of the Taiko chain. That let them withdraw USDC, USDT, crvUSD, ETH, TAIKO and other tokens that had never been sent across. The root problem was a signing key left in a public code repository plus a registration check that did not reject insecure enclaves; withdrawal limits kept more than $11 million in the bridge untouched. Taiko paused the bridge, refilled it 1:1 and made every user whole before reopening on July 2. The attacker sent about 810 ETH through Tornado Cash, while 1.99 million TAIKO was frozen at MEXC.
The attacker's main wallet used a helper contract to register rogue provers in one transaction, finalized fake checkpoints, and then withdrew tokens from the ERC20 Vault and ETH from the Bridge into a second wallet before mixing the ETH through Tornado Cash.
Attack flow
All 5 on-chain stepsThe attack in 3 steps. Step 1, Rogue provers: Main attacker wallet (attacker). Step 2, Drain: ERC20 Vault (source), Bridge (bridge), about $1.24M reached this step. Step 3, Launder: Consolidation wallet (hub), Tornado Cash and MEXC (exit), about $1.4M reached this step.
Step 1: Rogue provers
- AttackerMain attacker walletregisters two rogue proversposts fake checkpoints0x7506DeA0...e1B76a
Step 2: Drain
- SourceERC20 Vault675,761 USDC, 138,140 USDT, 156,832 crvUSD1.99M TAIKO, CRV, WETH, WBTC0x996282cA...4415Ab
- BridgeBridgeabout 296 ETH released278 ETH to the attacker0xd60247c6...Ddd8EC
Step 3: Launder
- HubConsolidation walletETH withdrawals plus swap proceedsfeeds the main wallet0xA9803508...38a846
- ExitTornado Cash and MEXCabout 810 ETH mixed1.99M TAIKO frozen at MEXC
Why each step worked
- Enclave signing key in a public repositorylet outsiders sign provers the contracts trusted
- Registration accepted insecure enclavesdebug mode was never rejected
- Trusted hardware alone could finalize statezero-knowledge proofs were optional
- Some tokens had no withdrawal limitmissing controllower-value tokens drained in full
- Pause took hours to executemissing controlfirst council proposals reverted; pause landed at 05:40 UTC
Full forensic detail
Step-by-step reconstruction, root cause, counterfactuals, remediation, and disclosure timeline.
Exploit anatomy
Root cause
This was an operational key exposure combined with a gap in how Taiko admitted provers, not a break in the bridge contracts or in the proving math, per Taiko Labs' post-mortem. Taiko's Ethereum contracts accepted proofs from approved SGX prover enclaves and identified trusted enclaves by the key that signed their software. That signing key had been committed to Taiko's public prover repository, and the registration check did not reject enclaves running in a debug mode that offers no confidentiality. Together these let an outsider register provers the contracts treated as Taiko's own. Taiko also allowed two SGX proofs to finalize a block, with zero-knowledge proofs optional, and it kept a permissionless proving fallback for liveness, so rogue provers were enough to finalize a fake version of the Taiko chain on Ethereum. The bridge then honored withdrawals backed by that fake state. Per-asset withdrawal limits capped the damage: about $1.75 million left, while more than $11 million in the Bridge and Vault never moved. Taiko has since rejected debug-mode enclaves at registration and, through its Unzen upgrade, made a zero-knowledge proof mandatory on every block.
Prevention analysis
The attack began with a signing key sitting in a public repository. Secret scanning on commits, and holding release signing keys in hardware, would have left the attacker with nothing to sign rogue provers with.
Even with the key exposed, checking the enclave's production flag would have blocked the rogue provers. Taiko deployed this fix on June 29.
With zero-knowledge proofs optional, a single trusted-hardware path could finalize state alone. Making a zero-knowledge proof mandatory, as Taiko's Unzen upgrade now does, removes that single point of failure.
Limits are why only about $1.75 million of more than $13 million left. Tokens without a limit were drained in full, so covering every asset would have shrunk the loss further.
Similar incidents
The bridge accepted an attestation it should have rejected and released funds on Ethereum-side state that never existed. Same class: a bridge that trusts a verifier step which an outsider managed to satisfy.
Forged guardian attestations were accepted by the bridge and released about $815,000. Same family: forged proof of a cross-chain event that never happened.
A rollup bridge paid out against a proof that should have failed. Same lesson: a rollup bridge is only as strong as the weakest path that can finalize state on Ethereum.
Remediation
Timeline
Get your protocol scored across 12 dimensions, or request ongoing coverage.