Babylon risk

TL;DR

Babylon's core claim survives scrutiny. Your BTC never leaves Bitcoin, and no Babylon party can move it. The staking output is a Taproot UTXO with key-path spending disabled by a NUMS internal key, and all three script paths begin with <StakerPk> OP_CHECKSIGVERIFY. There is no bridge, no wrapper, no custodian.

Two claims in our July report were wrong and are corrected here. Unbonding is not seven days: the chain parameter is 301 Bitcoin blocks, about 50 hours. And the covenant committee does not gate your exit. It signs your unbonding transaction at registration, before the stake is ever active, and those signatures are published on Babylon Genesis for anyone to read. We sampled 300 active delegations: every one already carried at least the 6-of-9 quorum on its unbonding transaction, and not one had requested to unbond. Babylon's own specification is blunt about it: "There's no way the covenant committee can act against the stakers, except rejecting their staking requests." The committee is a gate on entry, not on exit.

So the exit is unilateral and roughly two days. What remains is not custody risk. It is that rewards arrive in a token down 93% with three fifths of supply not yet circulating and no cap on it, that two finality providers both branded Kraken hold 79.6% of the stake that actually earns, and that the Genesis chain has needed eight security advisories in fifteen months, five of them high severity.

Checklist

Audits & contracts. Better covered than we credited in July. Six reports are published on Babylon's own documentation site and all six resolve: Coinspect on Phase-1 (2024) and Phase-2 (March 2025), Zellic on Phase-1 (April to May 2024) and on the Genesis chain (December 2024 to February 2025), plus Oak Security and Informal Systems on the Genesis V2 upgrade (both June 2025). Cantina ran a public competition in July 2024 titled Bitcoin Staking Scripts, which is precisely the layer that holds your coins. Immunefi has been live since 16 September 2024 at $500,000 maximum across 16 assets, page updated 27 July 2026. The caveats are real: Zellic's Genesis assessment returned 32 findings including seven critical, and OpenZeppelin's 2025 work was unpaid independent research rather than a paid engagement, found four on-chain issues plus problems in the Vigilante components, and stated plainly that it "did not cover Bitcoin-side staking scripts".

Admin control. The staking UTXO cannot be spent without the staker key, full stop. Governance sets the covenant set and quorum through MsgUpdateParams, whose authority field is documented in the proto as "the address of the governance account". That power is bounded in a way worth understanding: a Bitcoin script is fixed once funded, and every delegation records the parameter version it was created under, so a governance change binds only new stakes. The finality-provider set is permissionless to join and self-regulating: registration is on chain, up to max_active_finality_providers (60) can be active, and missing votes triggers a 600-second jailing rather than any loss of coins. The residual is that the covenant set is a governed parameter for future delegations, and that the mainnet committee's member identities are not published anywhere we could find.

Oracle. Nothing to report, which is the point. No price feed sits in the fund-loss path, there is no lending and no liquidation, and the trusted inputs are the Bitcoin light client and EOTS double-sign detection. Neither can move BTC.

Liquidity and exit. You hold a covenant-signed unbonding transaction from the moment your stake is verified. Broadcast it whenever you like, wait 301 blocks, then sweep the output with your key alone, because the unbonding output's timelock path is <StakerPK> OP_CHECKSIGVERIFY <TimelockBlocks> OP_CHECKSEQUENCEVERIFY. The cost is a 9,600 sat unbonding fee. The genuine constraints are that 50 hours is not instant, that no secondary market exists for a raw delegation, and that if you broadcast a staking transaction before registering it you can be left with only the 64,000-block timelock, about 444 days.

Yield: real vs emission. Rewards come in BABY. It trades at $0.0117, 93% below its April 2025 high, with a $50.3M market cap and a $127.7M fully diluted valuation. Only 4.30B of 10.92B tokens circulate, and there is no maximum supply: the token inflates around 8% a year, so the 10.92B itself keeps growing. This is emission paid on a Bitcoin principal.

Holder concentration. Measure it against the stake that actually carries voting power and earns, which is the 40,851 BTC behind the 32 active finality providers. Kraken02 holds 44.3% and Kraken holds 35.3%. That is 79.6% for one brand, and either provider alone exceeds the one-third share that can stall BTC-backed finality. The top three reach 88.0%. Against all 44,133 BTC registered, the same pair is 73.6%.

Track record. No fund-loss exploit: DefiLlama's hacks list runs to 621 entries and contains no Babylon and no Lombard. No slashing either: zero of 132 registered finality providers carry a non-zero slashed_btc_height. But the chain layer has been patched steadily. Eight GitHub advisories since May 2025, five rated high, including a validator-set bug that halts the chain (July 2025) and the BLS vote-extension panic disclosed 8 December 2025 and fixed in v4.2.0. Every one is a liveness or accounting fault on Babylon Genesis, none touches Bitcoin custody, and none was reported exploited. That is the right shape of failure, at an uncomfortable cadence for a two-year-old network holding billions.

Worst case

Separate principal from yield, because they fail differently.

Principal, protocol failure. Babylon Genesis stops producing blocks and the covenant disbands. Your BTC is fine. You already hold the signed unbonding transaction, so you broadcast it, wait 50 hours, and sweep. If you somehow never obtained covenant signatures, you wait out the remaining 64,000-block timelock, up to about 14.6 months, and sweep with your own key. Either way you lose time, not coins.

Principal, slashing. Your finality provider double-signs. A transaction you pre-signed at registration burns 0.1% of the stake to an OP_RETURN carrying "babylon", and the remainder returns to you after a 301-block timelock. On a large position that is a rounding error. On a small one it is not, because the slashing transaction carries a fee floor of 150,000 sat regardless of position size: at the 0.005 BTC minimum stake the burn is 500 sat while the fee floor alone is about 30% of your position. Size accordingly, and note that you cannot revoke that pre-signature once given.

Yield. BABY keeps bleeding against an uncapped supply and your realised return, measured in BTC, lands near zero or below.

The case that actually costs you is not Babylon's. Every guarantee above belongs to whoever holds the staker key. Hold LBTC or any other wrapper and that key is not yours, so you inherit the wrapper's custody scheme instead, and that needs its own assessment.

Bottom line

Caution, unchanged from 4 July, but for better reasons and with two of that report's claims retracted. Exit is faster than we said and is not committee-gated. The custody design is the strongest in BTC yield and we can point at the exact scripts that make it true. What keeps this off a clean rating is everything bolted around that core: a reward token down 93% with no supply cap, four fifths of the earning stake behind a single exchange brand, and a chain that has needed five high-severity patches in fifteen months. Stake directly rather than through a wrapper, pick a finality provider outside the Kraken pair, size for a 50-hour exit rather than an instant one, and treat BABY rewards as a lottery ticket rather than a yield.

Data appendix

• TVL: $2.6065B, verified from api.llama.fi/tvl/babylon-protocol. Slug note: our file slug is babylon, and api.llama.fi/tvl/babylon answers "Protocol not found", so our own scanner wiring needs checking. A re-read of the same endpoint hours later returned $2.6248B, so treat the figure as drifting about 1% intraday; the 2026-08-11 series point was $2.6214B.
• TVL history: 90d $4.0365B on 2026-05-14, so about -35%. ATH $7.1421B on 2025-10-07, so about -64%. Both from the babylon-protocol series. BTC-denominated history not pulled, so the split between price decline and real unstaking is refresh.
• TVL reconciliation: DefiLlama counts 40,964.30 BTC at about $64,076, which accounts for its entire figure. Babylon's staking API reports 44,132.91 BTC across 132 finality providers. The gap is not snapshot timing: DefiLlama counts only stake behind ACTIVE providers (40,851.31 BTC across 32 of them), excluding 3,002.80 BTC behind 78 inactive providers and 278.80 BTC behind 22 jailed ones.
• Audits: Coinspect Phase-1 (2024) and Phase-2 (2025-03); Zellic Phase-1 (2024-04 to 2024-05) and Genesis chain (2024-12 to 2025-02, 32 findings, 7 critical); Oak Security Genesis V2 (2025-06); Informal Systems Genesis V2 (2025-06). All six PDFs verified reachable on docs.babylonlabs.io this run. Cantina public competition, July 2024, scoped to the Bitcoin staking scripts. Immunefi live since 2024-09-16, $500,000 max, 16 assets, page updated 2026-07-27. OpenZeppelin 2025: independent research, not an engagement, four on-chain findings plus Vigilante issues, Bitcoin scripts explicitly out of scope.
• Custody model (verified from spec and chain): Taproot staking output, key path disabled via the BIP341 NUMS point. Three script paths: timelock (<StakerPK> OP_CHECKSIGVERIFY <TimelockBlocks> OP_CHECKSEQUENCEVERIFY), unbonding (staker plus covenant quorum), slashing (staker plus finality provider plus covenant quorum). Covenant co-signatures are "a pre-requisite for the stake's activation", and per the spec "There's no way the covenant committee can act against the stakers, except rejecting their staking requests."
• Pre-signed exit, measured: of 300 ACTIVE delegations sampled, all 300 carried a covenant unbonding signature list at or above the 6-signature quorum (64 at six, 79 at seven, 145 at eight, 12 at nine) and all 300 had no unbonding request recorded. Covenant signatures therefore exist before exit is requested, not in response to it. The delegation state machine confirms this: PENDING is "waiting for covenant signatures", VERIFIED is "has covenant signatures but is not yet included in the BTC chain".
• Key params (live, both nodes): covenant set 9 keys, covenant_quorum 6. Staking timelock fixed at 64,000 blocks, minimum equal to maximum, about 444 days. unbonding_time_blocks 301, about 50.2 hours. unbonding_fee_sat 9,600. slashing_rate 0.001. slashing_pk_script 6a07626162796c6f6e, an OP_RETURN carrying "babylon". min_slashing_tx_fee_sat 150,000. Stake bounds 0.005 to 5,000 BTC. Minimum finality-provider commission 3%. Delegations record params_version, so parameter changes bind new stakes only. MsgUpdateParams authority is documented in the proto as the governance account.
• Finality params: max_active_finality_providers 60, yet only 32 are active, so the cap is not what makes the other 100 inactive. jail_duration 600s, signed_blocks_window 10,000, min_signed_per_window 0.2.
• Concentration: against the 40,851.31 BTC active set, Kraken02 18,100.06 BTC (44.31%), Kraken 14,402.08 BTC (35.25%), Gate Earn 3,454.06 BTC (8.46%), Figment 2,086.97 BTC (5.11%). Kraken pair combined 79.56%, top three 88.02%. Against all 44,132.91 BTC registered the pair is 73.65%. Lombard x P2P.org holds 2,499.99 BTC but shows INACTIVE, which is 83% of all inactive stake.
• BABY token (CoinGecko, live): $0.01168787, market cap $50.27M, FDV $127.67M, circulating 4,300,276,396 of 10,920,708,226 total (39.4%), no maximum supply, ATH $0.166134 on 2025-04-12, now -92.96%. BTC reference price $64,075.
• Slashing history: zero of 132 registered finality providers show a non-zero slashed_btc_height. Twenty-two are jailed, holding 278.80 BTC between them; jailing is a 600-second liveness penalty, not a slashing.
• Incident record: eight GitHub security advisories on babylonlabs-io/babylon, five high severity, each with a patched release. Integer overflow causing chain halt and a MsgCommitPubRandList replay (both 2025-05-13, v1.1.0); fee-denomination chain crash (2025-06-28, 2.2.0); validator-set modification halting the chain (2025-07-07, 2.1.0); malformed vote extensions not rejected and BIP322 non-compliance (2025-11-24, 4.1.0); BLS vote-extension nil-BlockHash panic and costaking phantom-stake accounting (both 2025-12-08, 4.2.0). None reported exploited, none touching Bitcoin custody. Latest release v4.4.0 on 2026-08-06. DefiLlama hacks list: 621 entries, no Babylon, no Lombard.
• Developments: a16z crypto led a $15M raise announced 2026-01-07, structured as a BABY token purchase, funding Trustless BTCVaults. An Aave v4 Bitcoin-backed Spoke was targeted for April 2026; launch status refresh. A Ledger partnership was announced for vault access.
• Lombard boundary: Babylon's guarantee stops at the Bitcoin UTXO and belongs to whoever holds the staker key. LBTC holders do not hold that key, Lombard's consortium does, so Babylon's self-custody property does not pass through to LBTC and Lombard needs its own assessment.

Maintained monthly. Methodology: DeFi Research Instruction v2.