Threat model
How we compare to other vaults
How a vault works on the inside tells you surprisingly little about how it fails. A threat model looks at the edges instead: who can ask the vault for something, and what it takes to get a yes. For each product below, we lay out what an attacker needs to get that yes.
Most vaults protect a key with cryptography, but access is determined by a password or API credentials. An attacker rarely needs to break the cryptography, because it isn’t the weak point: stolen credentials open the door, and the vault does the rest. That’s why even impressive techniques such as MPC and DFC change nothing the way they’re used today. The SplitSecure Threshold Vault collapses cryptography and access control into one. The key doesn’t exist until enough of your Guardian Team approves, so a stolen credential has nothing to unlock. That is the core of what it does differently.
Some work has to run with nobody there to approve it. For that we also offer the SplitSecure Online Vault, in early access. It works like the other policy-based vaults, so it appears below with the same answer they get.
Passwords
Passwords are a legacy technology: a password has to be handed over in full every time it is used. We manage them so your team can leave its password manager, and then replace them, login by login, with a passkey, single sign-on or a certificate. See Privileged access.
This table is based on the public documentation of 1Password, Bitwarden and LastPass, and on each product’s default configuration. If a claim about a product is wrong, write to contact@splitsecure.com and we will correct it.
†1 With a stolen copy of the vault, the master password is enough to open it offline. 1Password also requires a Secret Key, which is created on your device and never sent to its servers.
†2 Your Guardian Team is the set of devices you choose to approve requests. Each one holds shares of the key on attested hardware that you own: Apple devices with App Attest, Android devices with StrongBox, AWS Nitro Enclaves, AMD SEV-SNP, Intel TDX or Google Cloud Confidential Space. SplitSecure holds no shares. One approval can cover a batch of passwords.
†3 A released password is always encrypted for the device that asked, with a key that the device uses once and then deletes. This gives each release forward secrecy.
Secrets
This table is based on the public documentation of HashiCorp, AWS and Akeyless, and on each product’s default configuration. If a claim about a product is wrong, write to contact@splitsecure.com and we will correct it.
†1 Akeyless lets a customer hold one fragment on its own gateway. Akeyless states that such an operation completes when identity and policy checks pass. The fragment protects against the vendor, and not against a stolen access token.
†2 Your Guardian Team is the set of devices you choose to approve requests. Each one holds shares of the key on attested hardware that you own: Apple devices with App Attest, Android devices with StrongBox, AWS Nitro Enclaves, AMD SEV-SNP, Intel TDX or Google Cloud Confidential Space. SplitSecure holds no shares. One approval can cover a batch of secrets.
†3 A released secret is always encrypted for the device that asked, with a key that the device uses once and then deletes. This gives each release forward secrecy.
Privileged access
This table is based on the public documentation of CyberArk, BeyondTrust, HashiCorp and Teleport, and on each product’s default configuration. If a claim about a product is wrong, write to contact@splitsecure.com and we will correct it.
†1 These products can add an approval step. The same server enforces it, so an attacker who takes the server skips it.
†2 Your Guardian Team is the set of devices you choose to approve requests. Each one holds shares of the key on attested hardware that you own: Apple devices with App Attest, Android devices with StrongBox, AWS Nitro Enclaves, AMD SEV-SNP, Intel TDX or Google Cloud Confidential Space. SplitSecure holds no shares. One approval can cover a batch of requests.
Signing keys
This table is based on the public documentation of HashiCorp and Akeyless, and on each product’s default configuration. If a claim about a product is wrong, write to contact@splitsecure.com and we will correct it.
†1 HashiCorp Vault can add an approval step with control groups, a feature of Vault Enterprise. The same server enforces it.
†2 Akeyless lets a customer hold one fragment on its own gateway. Akeyless states that such an operation completes when identity and policy checks pass. The fragment protects against the vendor, and not against a stolen access token.
†3 Your Guardian Team is the set of devices you choose to approve requests. Each one holds shares of the key on attested hardware that you own: Apple devices with App Attest, Android devices with StrongBox, AWS Nitro Enclaves, AMD SEV-SNP, Intel TDX or Google Cloud Confidential Space. SplitSecure holds no shares. One approval can cover a batch of signatures.
Digital assets
This table is based on the public documentation of Ledger, Trezor, Safe, Squads, Fireblocks and Dfns, and on each product’s default configuration. If a claim about a product is wrong, write to contact@splitsecure.com and we will correct it.
†1 The approvals are recorded inside the transaction or as transactions of their own, depending on the wallet. A chain with signature aggregation, such as BLS or Schnorr, avoids the extra cost.
†2 Fireblocks authenticates an API call with an API key and a request that the caller signs with its private key. Both are on the caller’s server.
†3 Fireblocks and Dfns can require approvers for a transaction. The policy engine enforces that rule too.
†4 Your Guardian Team is the set of devices you choose to approve requests. Each one holds shares of the key on attested hardware that you own: Apple devices with App Attest, Android devices with StrongBox, AWS Nitro Enclaves, AMD SEV-SNP, Intel TDX or Google Cloud Confidential Space. SplitSecure holds no shares. One approval can cover a batch of transactions.