Most large losses on distributed ledgers do not come from breaking the cryptography or the consensus protocol, but from the infrastructure around them: validator and bridge keys, exposed nodes, and the people and machines that sign. This guide covers the layers below and beside smart contracts: consensus and finality, cross-chain bridges and their trust models, validator and signer key custody, node and RPC hardening, and permissioned ledgers such as Hyperledger Fabric. It is written for security engineers, infrastructure teams and risk owners at exchanges, custodians, protocol teams, validators and enterprises running private ledgers. Contract code is covered in the Smart Contract Security Guide and the overall programme in the Web3 and Crypto Security Guide; this page goes deeper on the ledger infrastructure and does not repeat them.
Key takeaways
- Know your trust assumptions. Every ledger, bridge and oracle rests on a stated or unstated assumption about who can sign, how many must collude and what happens if they do. Write it down.
- Consensus thresholds are security boundaries. On proof of stake Ethereum, one third of stake can delay finality and two thirds can finalise at will; on smaller chains these levels may be affordable.
- Bridges concentrate risk. Externally verified bridges are only as strong as their signer set, and ethereum.org notes bridges account for three of the biggest DeFi hacks.
- Validator and signer keys are non-human identities. They need inventory, custody, rotation and revocation like any privileged machine credential, and their use must be monitored.
- Nodes are not hardened services. Client RPC interfaces are not built for hostile traffic; keep them private and put signing on separate hardware.
Ledger types and their security models
NIST's Blockchain Technology Overview (NISTIR 8202) describes blockchains as distributed ledgers that are tamper evident and tamper resistant, usually without a central authority, and lists consensus models including proof of work, proof of stake, round robin, proof of authority, proof of identity and proof of elapsed time. ISO/TC 307 maintains the vocabulary standard, now ISO 22739:2024 (the 2020 edition was withdrawn in January 2024). The security questions differ by ledger type:
| Ledger type | Who validates | Main security dependency | Typical failure |
|---|---|---|---|
| Public proof of work | Anyone with hash power | Cost of majority hash power | Deep reorganisations on chains with low hash rate |
| Public proof of stake | Anyone with stake, as validators | Cost and distribution of stake; slashing; client diversity | Finality delay, reorgs, validator key theft or slashing |
| Proof of authority or small validator sets (including many sidechains) | A named, small set of validators | Security of each validator's keys and independence of operators | Majority of validator keys compromised through one organisation |
| Permissioned enterprise ledger (for example Hyperledger Fabric) | Member organisations' nodes | Membership PKI, policies and node security | Compromised admin or CA, weak endorsement policy |
For decision makers, the key point is that a "decentralised" label does not tell you the trust model. A chain with seven validators run by three companies has a trust model closer to a consortium than to a public ledger.
Consensus and finality attacks
Ethereum's documentation on proof of stake attacks and defences sets out what different shares of staked ETH allow an attacker to do. These figures are specific to Ethereum's Gasper design, but the shape applies to many BFT-style chains:
| Attacker share of stake | What becomes possible | Built-in defence |
|---|---|---|
| About 33% | Prevent finality by withholding attestations | Inactivity leak drains non-participating stake until honest validators regain more than two thirds |
| 34% | Double finality, with some control of message timing | Slashing; the attacker loses its stake to the correlation penalty |
| About 50% | Keep two forks alive and block finality | Inactivity leak eventually finalises both; social recovery is the fallback |
| Over 50% | Control fork choice: short reorgs, censorship, MEV extraction | Social layer response |
| 66% or more | Finalise its preferred chain at will | Social layer, including a coordinated fork |
Smaller attacks need no threshold: ex ante reorgs by withholding blocks, balancing and bouncing attacks that exploit vote timing, and denial of service against a known upcoming block proposer. Long-range attacks are prevented by finality and weak subjectivity checkpoints that new nodes start from. Slashing applies to three offences: double proposal, double vote and surround vote. A slashed validator is forcibly exited over 36 days, with a correlation penalty at the midpoint that grows with how many other validators were slashed at the same time.
What this means for defenders:
- Exchanges and bridges: set confirmation or finality requirements per chain from its economic security, not a single number for all chains. Wait for finality on chains that offer it before crediting large deposits or releasing bridged funds.
- Validators: correlated failure is what the correlation penalty punishes, so avoid running all keys on one client, one cloud region or one signer.
- Risk owners: on chains where majority stake or hash power is cheap, treat deep reorganisations as a credible event and plan for it.
Bridge security and trust models
A bridge locks or burns assets on one chain and mints or releases them on another, based on some proof that the first event happened. ethereum.org groups bridges into native bridges, validator or oracle based bridges, generalised message passing bridges and liquidity networks, and distinguishes trusted (externally verified) from trustless designs:
| Verification model | Who you trust | Strength | Weakness | Controls to look for |
|---|---|---|---|---|
| Multisig or MPC signer set | A threshold of external signers | Fast, cheap, connects many chains | Collusion or theft of a threshold of keys | Independent operators, high threshold, HSM keys, outflow limits, monitoring |
| Oracle network | The oracle operators | Flexible | Same as signer set, plus oracle manipulation | Diverse operators, rate limits |
| Light client (for example IBC) | The source chain's own consensus | No new trust assumption | Limited connectivity, client implementation bugs | Audited light client, upgrade governance |
| Optimistic with fraud proofs | At least one honest watcher during the challenge window | Lower cost | Slower; fails if watchers are absent or the contract has a bug | Funded independent watchers, adequate challenge period |
| Liquidity network (atomic swaps) | The underlying chains | Strong security | Cannot pass arbitrary messages | Timeout handling |
ethereum.org lists the risks as smart contract risk, systemic financial risk from wrapped assets and counterparty risk in trusted designs, and notes that behaviour under congestion, network attacks or state rollbacks is not well understood. Operational controls matter as much as the design:
- Signer independence: no single organisation, cloud account or person should be able to reach a threshold of bridge keys.
- Rate limits and delays: cap outflows per period and delay large withdrawals so that monitoring can act before funds leave.
- Outflow monitoring: alert on unusual withdrawal size or frequency. See the Blockchain Analytics and On-Chain Monitoring Guide.
- Pause capability: a tested emergency pause, held by a separate key set, with clear authority to use it.
- Revocation discipline: remove temporary signing delegations as soon as they are no longer needed.
Validator, signer and custody keys
Ledger infrastructure runs on keys that act without a human at the moment of use: validator signing keys, bridge signer keys, oracle keys, hot wallet keys, deployer and admin keys, and the CI and cloud credentials that can reach them. Treat each as a non-human identity with an owner, a purpose, a location, a rotation plan and a revocation path. See the Crypto Exchange Identity and Key Security Guide.
| Key | Impact if stolen | Custody approach | Specific hazard |
|---|---|---|---|
| Validator signing key | Slashing, missed duties; with many keys, consensus influence | Remote signer or HSM, separate from the beacon or consensus node | Running the same key in two places causes slashable double signing |
| Staking withdrawal credentials | Theft of staked funds on exit | Offline, ideally a contract or multisig address | Kept on the same host as signing keys |
| Bridge or oracle signer key | Unauthorised release of bridged funds | HSM or MPC per operator, distinct operators | Several keys reachable from one corporate network |
| Contract admin or upgrade key | Arbitrary contract changes | Multisig with timelock | Signers approving what a compromised interface shows them |
| Hot wallet key | Direct theft up to wallet balance | HSM or MPC, policy engine, low balance | Balance far above operational need |
| Fabric admin and CA keys | Enrol identities, change channel configuration | HSM via PKCS#11, offline root CA | Admin certificates distributed widely |
When moving a validator between clients or hosts, use the slashing protection interchange format (EIP-3076, in Last Call status when checked) so the new client knows what the key has already signed, and export or import only while the validator client is stopped. Never run a "hot spare" with the same key. For threshold signing designs, see the MPC Wallet and Threshold Signing Security Guide, and for ceremonies the Cryptographic Key Ceremony Template.
Node and RPC hardening
Node clients are network software with a large attack surface. The Geth security documentation gives concrete rules that generalise to other clients:
- Keep RPC private: block the JSON-RPC port (8545 by default) except from explicitly trusted machines. HTTP-RPC, WebSocket, GraphQL and the beacon API are not designed to withstand hostile clients; exposure risks crashes, resource starvation and spurious signing requests.
- Open only peer ports: allow TCP and UDP 30303 (or your custom peer port) for peering and discovery.
- Limit namespaces: expose only the API namespaces each consumer needs; signing risk depends on which are enabled.
- Front any necessary exposure: if an endpoint must be reachable, put a proxy with TLS, request filtering, rate limiting, logging and monitoring in front of it.
- Separate signing from the node: use a dedicated signer (Geth recommends Clef) on separate hardware, so the node never holds keys.
- Verify binaries: check downloads against published SHA256 hashes, for both execution and consensus clients.
Third-party RPC providers are a supplier dependency: their API keys are secrets, their responses determine what your systems believe the chain state is, and an outage stops your services. Use more than one provider or your own node for critical reads, and treat RPC keys like any other API key. See the API Key Management Guide.
Permissioned ledgers: Hyperledger Fabric
In a permissioned ledger the cryptography is the same but trust moves to membership and policy. Fabric's security model gives every peer, orderer, application and administrator an X.509 identity issued by a certificate authority, with a Membership Service Provider (MSP) defining which identities are valid for each organisation. Practical controls:
- Endorsement policies: require endorsement from more than one organisation for any chaincode that moves value or changes shared records, so a single compromised member cannot write alone.
- Lifecycle and channel policies: set the LifecycleEndorsement policy and channel configuration modification policies so no single organisation can deploy chaincode or change the consenter set.
- Node OU roles: use client and admin roles in policies and keep admin certificates few and short-lived.
- Mutual TLS: enable mutual TLS between nodes; the operations service relies on mutual TLS with client certificates.
- HSM-backed keys: Fabric supports PKCS#11 HSMs so node private keys never leave the device.
- Client applications: client identities can only submit transactions and query; enforcement happens on nodes, not in SDKs, so do not rely on client-side checks.
See the Enterprise PKI Design and CA Hierarchy Guide for the CA design underneath.
Implementing a ledger infrastructure security programme
- Map trust assumptions: for each chain, bridge, oracle and RPC provider you depend on, record the validator or signer set, threshold, operators and upgrade authority.
- Inventory keys and credentials: every validator, signer, admin and hot wallet key, plus the cloud, CI and developer credentials that can reach them.
- Separate and harden: move signing to HSMs, MPC or remote signers; isolate signer hosts from corporate networks; make RPC private.
- Set per-chain policies: finality or confirmation depth, bridge outflow limits and withdrawal delays.
- Test on a testnet first: validator migrations with slashing protection, key rotations, pause procedures and failover between RPC providers. Roll back by restoring the previous signer configuration, never by running two signers.
- Exercise incidents: bridge key compromise, consensus reorg, validator slashing and RPC provider outage. See the Web3 Incident Response Playbook.
Small versus large organisations: a small team staking or integrating a few chains is usually better served by a reputable staking or custody provider with documented controls, plus its own monitoring, than by running validators. A large exchange, bridge operator or custodian should run its own nodes for critical reads, operate independent signer infrastructure per key set, and have a 24 by 7 monitoring and pause capability.
Detection and measurement
- On chain: bridge and hot wallet outflows against baselines, admin or upgrade transactions, changes to signer sets, and reorg depth on chains you credit.
- Validators: missed attestations or proposals, slashing events, and any signature from a key outside its expected signer host.
- Infrastructure: access to signer hosts, HSM audit logs, RPC request patterns and namespace use, and new peers or unexpected client versions.
Metrics:
- Time to detect an abnormal bridge or hot wallet outflow, measured in exercises.
- Percentage of signing keys in HSM, MPC or remote signer custody.
- Maximum number of bridge or validator keys reachable from a single organisation, network or cloud account.
- Number of RPC endpoints reachable from the internet without a filtering proxy (target: zero).
- Days since the last tested pause or validator migration procedure.
Framework mapping
| Area | MITRE ATT&CK | NIST CSF 2.0 | NIST SP 800-53 Rev. 5 | ISO/IEC 27001:2022 Annex A |
|---|---|---|---|---|
| Validator and signer key theft | T1552.004 Private Keys | PR.DS-01 | SC-12, SC-13 | A.8.24 |
| Social engineering of operators | T1566.001 Spearphishing Attachment | PR.AT-01 | AT-2 | A.6.3 |
| Compromised front end or client software | T1195.002 Compromise Software Supply Chain | GV.SC-07 | SR-3, SA-11 | A.5.21 |
| Abuse of signing delegations and accounts | T1078 Valid Accounts | PR.AA-05 | AC-6 | A.5.18, A.8.2 |
| Exposed nodes and RPC | T1190 Exploit Public-Facing Application; T1498 Network Denial of Service | PR.IR-01 | SC-7, CM-7 | A.8.20, A.8.22 |
| Theft of funds | T1657 Financial Theft | DE.CM-09, RS.MI-01 | SI-4, IR-4 | A.8.16, A.5.26 |
Common failure patterns
- A threshold reachable through one organisation: in the Ronin bridge breach, Sky Mavis's postmortem reports the attacker controlled five of nine validator keys, four of them Sky Mavis keys reached after a spear-phishing compromise of an employee, and took 173,600 ETH and 25.5 million USDC on 23 March 2022. The FBI attributed the theft to the Lazarus Group and APT38. See the Lazarus Group Threat Actor Profile.
- Delegations never revoked: the fifth Ronin key came from the Axie DAO, which had allowlisted Sky Mavis to sign on its behalf in November 2021; the arrangement ended in December 2021 but the access was never revoked and was reached through a gas-free RPC node.
- No outflow monitoring: the same theft was discovered six days later, on 29 March 2022, and the postmortem states there was no tracking system for large bridge outflows.
- Trusting the signing interface: in the Bybit hack, attackers used a developer's hijacked AWS session tokens to alter the multisig web interface, so genuine signers approved a different transaction than the one they saw, and about $1.5 billion was taken.
- Exposed RPC with signing enabled: node APIs reachable from the internet with account or signing namespaces enabled.
- Duplicate validators: "high availability" setups that run the same validator key twice and get slashed for double signing.
Practitioner checklist
- Document the trust model, signer set, threshold and upgrade authority for every chain and bridge you depend on.
- Inventory every validator, signer, admin and hot wallet key with an owner and custody location.
- Ensure no single organisation, network or cloud account can reach a signing threshold.
- Keep signing keys in HSM, MPC or remote signers, separate from nodes.
- Block public access to node RPC, GraphQL and beacon APIs; open only peer ports.
- Set per-chain finality or confirmation rules and bridge outflow limits with delays.
- Monitor outflows, signer set changes and validator duties, with a tested pause.
- Use slashing protection data on every validator migration and never run duplicate keys.
- Revoke temporary signing delegations and RPC allowlists on a fixed review cycle.
Standards and references
- NIST: NISTIR 8202 Blockchain Technology Overview
- ISO: ISO 22739 Blockchain and distributed ledger technologies, Vocabulary
- Ethereum Foundation: Proof of stake attack and defense
- Ethereum Foundation: Bridges
- Ethereum Improvement Proposals: EIP-3076 Slashing Protection Interchange Format
- Go Ethereum: Security
- Hyperledger Fabric: Security Model
- Ronin: Back to Building, Ronin Security Breach Postmortem
Related NHI Mgmt Group resources: Web3 and Crypto Security Guide · Smart Contract Security Guide · Crypto Exchange Identity and Key Security Guide · MPC Wallet and Threshold Signing Security Guide · Web3 Incident Response Playbook