Klever-Go CVE-2026-82407

Exploitation requires the ability to submit a CreateValidatorContract (Register) or ValidatorConfig update (UpdateValidator) transaction on a Klever-Go node running a version prior to 1.7.20 - i.e. an actor with validator-registration capability (stake/gas), hence PR:L, not fully unauthenticated. The submitted BLSPublicKey must be malformed (not a valid, nonzero, prime-order-subgroup BLS12-381 G2 point). Crucially, impact is not immediate: the malformed-key validator must become ELIGIBLE and then be probabilistically SELECTED into a consensus group before MultiSigner.Reset fails (the AT:P gate). Genesis validators are NOT exploitable because the genesis path already runs CheckPublicKeyValid. The worst-case full halt only occurs when the consensus group equals the eligible validator set; larger validator sets suffer throughput degradation and missed rounds rather than a complete stop. The fix is additionally gated behind the FixAuditChangesV3 fork for deterministic historical block reprocessing.

This is a genuine but conditional availability-only risk, not a high-CVSS-but-hollow issue. The vendor CVSS 4.0 vector (AV:N/AC:L/AT:P/PR:L/UI:N/VA:H/SA:H, 7.0) correctly captures the essence: network-reachable, low-privilege (any actor able to submit a Register or UpdateValidator transaction, i.e. stake/gas to become a validator), no user interaction, pure availability impact (VA:H/SA:H, no confidentiality or integrity). The critical nuance is AT:P (Attack Requirements: Present) - exploitation is probabilistic because the malformed-key validator must actually be SELECTED into a consensus group before MultiSigner.Reset fails; this is a real gating condition that a raw 3.1 vector would express as AC:H. My independent 3.1 assessment is ~6.5 (AV:N/AC:H/PR:L/UI:N/S:C/C:N/I:N/A:H), where S:C reflects one validator's bad key degrading the whole network's consensus throughput. Impact scales from missed rounds/throughput degradation up to a COMPLETE HALT when the consensus group equals the eligible validator set. No EPSS score, no CISA KEV listing, and no weaponized public exploit are present; a security regression test ships with the fix but that is not an attacker POC. Signals are consistent - moderate severity, availability-only, self-reported by the vendor - so this is a real priority for anyone running a Klever-Go validator, but the niche footprint and probabilistic selection gate keep real-world urgency moderate rather than critical.

An attacker with enough stake/access to submit validator transactions registers (or updates) a validator supplying a BLSPublicKey that is not a valid BLS12-381 G2 curve point. Because pre-fork Register/UpdateValidator skip the curve/subgroup/nonzero checks (only genesis validation ran CheckPublicKeyValid), the malformed key is accepted; once that validator becomes eligible and is selected into a consensus group, MultiSigner.Reset cannot deserialize the group key and cancels the slot, causing repeated missed rounds and, if the consensus group equals the eligible set, a full network halt. No public exploit is identified at time of analysis.