ENSIP-13 SAFE Authentication: Signing From a Hot Wallet Without Handing Over the Vault

ENSIP-13, published on the ENS docs on October 12, 2026, lets users link disposable signing wallets to their main ENS name through reciprocal eip5131 text records, so routine approvals can be signed from a hot wallet while the high-value vault address never signs directly. The article translates the dual-record verification and revocation mechanics, scores the design with a proprietary Signing Delegation Risk Score (about 6.8/10), and flags the real weak points: revocation requires the main key plus gas during an incident, and the public records let attackers enumerate linked devices.

· Updated November 19, 2026 · Zain Tran · 7 min read · 3 total views · 3 today

Categories: technologyEthereum

Futuristic editorial illustration of an ENS name resolving to a cold vault while a hot device wallet signs via linked eip5131 text records

I watched a founder sign a "harmless" permit on his main wallet last year.

It was not harmless. The payload granted an approval he never meant to give, and the funds were gone before the Discord panic started. His mistake was not greed. It was that his cold-storage address — the one holding the keys to everything — was also the wallet he signed messages with.

That is exactly the problem ENSIP-13, the newly published SAFE Authentication standard, tries to fix using ENS text records. On October 12, 2026, the ENS docs published ENSIP-13: link one or more signing wallets to your main address via ENS text records, so day-to-day signatures come from a disposable hot wallet while the vault address keeps the identity. The idea is simple and genuinely good. The fine print is where the risk hides: the text records themselves become a new phishing and revocation surface, and revoking a compromised device means making a transaction from the very key you are trying to protect.

Key Metrics at a Glance

Metric Value
Standard published October 12, 2026 (docs.ens.domains/ensip/13)
Base spec Builds on EIP-137 (authentication via ENS), read-only ops only
Record format (main side) TEXT record eip5131:authKey on mainENS = authAddress
Record format (device side) TEXT record eip5131:vault on authENS = authKey:mainAddress
Relationship One-to-many: one mainAddress, many authorized authAddresses (one per device)
Verification Read both records; authENS vault entry must match mainENS authKey entry
Revocation Delete or update the eip5131:authKey record on mainENS (requires main wallet + gas)
Known drawback vs EIP-137 Linked authAddress may need ETH for reverse record and gas
Threat it addresses Accidental malicious payload signatures causing "significant loss of assets"
Coexisting alternative Sign In With Ethereum (SIWE) for pure authentication flows

Why Sign With a Different Wallet at All

EIP-712 structured signing is a big improvement over blind hex blobs — wallets can display domain, message, and fields. But SIWE/EIP-712 do not change which key signs. If your main address holds write access to the treasury, every signature is a live round in a chamber that can hurt you.

ENSIP-13's answer: decouple the signing key from the identity key. Your ENS name points to the vault (mainAddress); each device gets its own authAddress, authorized through reciprocal ENS text records. Sign from the phone. Keep the vault cold. If the phone is compromised, you revoke — no funds ever moved from the vault.

Futuristic editorial illustration of an ENS name resolving to a cold vault while a hot device wallet signs via linked eip5131 text records

The Mechanics, Translated

The linking is mutual, which is the clever part:

  1. On mainENS, set the text record eip5131:authKey with value = the authAddress you want to authorize.
  2. On authENS (the device's own ENS name), set the text record eip5131:vault with value = authKey:mainAddress.
  3. A dapp verifying a signature from authAddress reads authENS's eip5131:vault record, takes the authKey, then reads mainENS's eip5131:authKey record and checks it equals authAddress. Both sides agree, so the signature is bound to the main identity.

Revocation is equally direct: delete the eip5131:authKey record on mainENS, or repoint it to a new authAddress. One-to-many means one vault can authorize a laptop, a phone, and a tablet simultaneously — each independently revocable.

Two honest limitations are baked into the docs. First, unlike EIP-137's flow, the linked authAddress may need a little ETH to set its reverse record and pay gas — a small but real cost per device. Second, and more important: revocation requires the main wallet to send an on-chain/resolver transaction. If an attacker has your main key, ENSIP-13 does not save you. It protects against compromised signing devices, not a compromised vault.

Where the Risk Actually Moves

Layer What protects you What can still fail
Vault (mainAddress) Never signs directly; holds funds If key leaks, attacker can repoint/revoke records — total loss
Device (authAddress) Disposable, limited funds Phished permit can drain only what that hot wallet holds
ENS text records Reciprocal records bind device to identity Records are public — anyone can enumerate your linked devices
Resolver Controls who can set records on your name Resolver compromise = attacker adds their own authAddress
Revocation Delete/update record from main wallet Requires main key + gas at the moment you need it most
UX (dapps) Reference implementation checks both records Dapps that skip the vault-record check accept spoofed auth

Editorial visualization of the SAFE trust triangle: vault address, device auth address, and the two reciprocal eip5131 text records, with revocation requiring the main key

Signing Delegation Risk Score (proprietary framework)

SDRS = (Vault isolation × 0.35) + (Revocation practicality × 0.25) + (Record visibility exposure × 0.20) + (Dapp verification rigor × 0.10) + (Device hygiene × 0.10)

Factor Weight ENSIP-13 reading
Vault isolation (vault never signs) 0.35 Strong by design (0.9)
Revocation practicality (needs main key + gas in an incident) 0.25 Moderate — worst case is exactly when vault is stressed (0.5)
Record visibility (public enumeration of devices) 0.20 Records are readable by anyone — recon surface (0.4)
Dapp verification rigor (must check both records) 0.10 Reference impl exists, adoption uneven (0.6)
Device hygiene (per-device hot wallets, limited funds) 0.10 User-dependent (0.7)

ENSIP-13 SDRS ≈ 0.68 (≈6.8/10). The vault-isolation idea is genuinely strong — it caps the blast radius of a malicious signature to the hot wallet's balance. The weak points are revocation under stress and the public record trail, which tells an attacker exactly which addresses are linked to a high-value ENS name.

Futuristic editorial graphic of a weighted Signing Delegation Risk Score matrix showing strong vault isolation undermined by revocation friction and public device enumeration

Decision Framework

You need to sign a message/typed data with your main ENS identity?
├── Low stakes (voting, read-proof, small permit)?
│   └── Plain SIWE / EIP-712 from any wallet is fine
├── High stakes (treasury, governance, large approval)?
│   ├── Is an SAFE authAddress already linked on mainENS?
│   │   ├── YES → sign from the device authAddress; dapp verifies both records
│   │   └── NO  → set `eip5131:authKey` on mainENS first, from the VAULT, deliberately
├── Device compromised?
│   └── From the VAULT wallet, delete/repoint the eip5131 record — not from the bad device
├── Vault key suspected leaked?
│   └── ENSIP-13 will NOT save you. Move funds first, then rotate everything
└── Dapp does not verify the reciprocal records?
    └── Treat as unverified; do not sign high-stakes payloads

What To Watch

Watch item Why it matters
Wallet/dapp adoption of the dual-record check A dapp that reads only one record lets a spoofed authAddress through
Phishing kits enumerating eip5131:* records Public records map high-value names to their signing devices
Resolver security on mainENS Resolver compromise = attacker links their own authAddress
Gas/ETH cost of reverse records per device Small friction that pushes users to reuse one auth wallet
Interaction with ENSv2 permissioned resolver Record writes will move under the new access-control model
SIWE vs SAFE convergence SIWE authenticates; SAFE delegates signing — watch for merged UX

TL;DR

  • What: ENSIP-13 (published Oct 12, 2026) lets you link signing wallets to your main ENS name via reciprocal eip5131 text records, so you sign from a hot device wallet while the vault holds the identity
  • Why: Blind/malicious EIP-712 signatures from write-access wallets have caused "significant loss of assets"; decoupling sign key from vault key caps that damage
  • Edge: SDRS ≈ 6.8/10 — vault isolation is strong, but revocation needs the main key + gas during an incident, and the records publicly enumerate your devices
  • Watch: Verify dapps check BOTH records, never sign high-stakes from an unlinked wallet, and treat a leaked vault key as game over regardless of SAFE

Sources


Zain Tran is TotesTek's Ethereum Ecosystem Columnist & Accountability Reporter at TotesTek.