← All Reports

GHO — Risk Report

Moderate risk · 5.5/10

What this actually is

GHO is Aave DAO’s overcollateralized stablecoin. It is minted by a set of authorised facilitators, each with a governance-set cap — a “bucket” — that bounds how much GHO it may issue. Supply is 699,000,000.00 across eight facilitator levels.

⚠️ Two things dominate this report and neither is about collateral quality. The peg’s defence mechanism is wound down (axis 1), and 44.3% of supply is minted by a contract whose authorised callers cannot be listed from chain state (axis 5).

1 · Stability — 5.5

Reference: a $1 par target. ⚠️ The mechanism that would defend it is wound down.

Both of Aave’s mainnet GHO Stability Modules are seized — first verified on-chain 2026-08-13, re-verified unchanged 2026-08-30:

0x0d8eFfC11dF3F229AA1EA0509BC9DFa632A13578   getIsSeized() TRUE   getIsFrozen() FALSE
   GHO 0.0 · exposureCap 0 · UNDERLYING_ASSET() 0xA0b86991c6218b36c1d19D4a2e9Eb0cE3606eB48 = USDC
0x686F8D21520f4ecEc7ba577be08354F4d1EB8262   getIsSeized() TRUE   getIsFrozen() FALSE
   GHO 0.0 · exposureCap 0 · UNDERLYING_ASSET() 0xdAC17F958D2ee523a2206206994597C13D831ec7 = USDT
GHO PegKeeper 0x53876B157DeCf04389eEd66c7C29d73863f8C50b   debt() 0.00

⚠️ Note how each module’s identity is established: by reading UNDERLYING_ASSET(), not by trusting its name. Same discipline as axis 5 below, and for the same reason — on this asset, labels have already proven the least reliable thing on the contract.

⚠️ seized = true alongside frozen = false is the signature of a wind-down, not a pause pending restart. Freezing halts a module you intend to restart; seizure is the GSM’s terminal emergency action. Neither module can swap GHO against USDC or USDT, so the mechanism that would arbitrage a sub-$1 GHO back toward par is not available.

Unchanged since 2026-08-13 is the more useful fact than either reading alone. A module seized today could be mid-incident; one that has held seized, unfrozen, empty and capped at zero for weeks is behaving like a decommissioned component — which makes this an observation rather than an inference.

And Curve’s Emergency DAO set the GHO PegKeeper debt ceiling to 0, so that keeper can neither mint nor burn and contributes nothing in either direction. ⚠️ What has not stopped: the GHO/crvUSD pool still exists and still provides DEX depth. What ended is its role as a defence mechanism, not its role as liquidity — see axis 3.

The two facts connect legibly: with no functioning GSM there is no reliable route from sub-par GHO back to $1, so a PegKeeper accumulating GHO would have been holding something it could not exit. The reported trigger for the GSM impairment is the April 2026 rsETH bridge exploit.

⚠️ Sourcing, because these have different standing. The seizure states and the zero ceiling are direct on-chain reads. The Curve governance action itself is reported rather than read — the Emergency DAO decision, its date and ID remain unverified, since gov.curve.finance returns HTTP 403 from two separate environments. See the crvUSD report.

2 · Backing — 6.0

GHO is overcollateralized and minted against Aave’s collateral basket, with each facilitator bounded by a governance-set bucket. The reconciliation works: eight facilitator levels sum to 699,000,000.00 against a totalSupply of 699,000,000.00 — zero delta.

⚠️ Supply is frozen, not merely slow, and every level is identical to the 2026-08-13 read. Not approximately — exactly.

2026-08-13   699,000,000.00   (the preceding week had run +7.7%)
2026-08-30   699,000,000.00   +0.0%
2026-09-08   699,000,000.00   +0.0% — every facilitator level identical to the wei

A growth rate that went from +7.7% in a week to exactly zero, and has held there since 2026-08-13, is a state change rather than a slowdown. 65.8% of supply sits at hard caps, so a substantial part of the float is structurally unable to expand without governance raising ceilings.

⚠️ Which direction the cap binds decides how to read that 65.8%. A facilitator at capacity cannot mint. So the leg the caps constrain is the defence against GHO trading above par — the arbitrage that mints fresh GHO into a premium and sells it back toward $1. GHO trades at $0.998702, which is 0.13% below par, so that is not the leg under load today.

Below par the correcting trade is a different one: buy GHO under $1 and repay debt at face value. It does not touch a facilitator bucket and works whatever the ceilings say. ⚠️ So “65.8% at hard caps” is not an active peg failure — and a reader meeting that figure next to “no functioning GSM” will reasonably conclude that it is. It is a latent, one-sided constraint, and clearing it needs a governance vote: the vote is the exposure, not the cap.

⚠️ That is not a reassurance, because both legs are impaired and only one of them is being tested. Upward correction is throttled by the caps. Downward correction lost its reliable route when both Stability Modules were seized, leaving borrower repayment — which depends on there being borrowers who want to repay. The leg currently under load is the impaired one without a mechanism behind it.

⚠️ And the caps cut both ways for a holder, which is why they do not move this axis alone. Caps that bind are a constraint on uncontrolled issuance, and equally a constraint on the asset’s ability to meet demand.

3 · Liquidity & Exit — 6.0

Both exit paths, and the axis takes the worse one. ⚠️ The binding leg here is the PRIMARY route, and it closed — stated explicitly so the axis does not read as a comment on depth.

Primary — the fixed-price swap is gone. A GSM let a holder swap GHO against USDC or USDT at a fixed price. Both modules are seized with exposureCap at zero (axis 1), so that route is not available at any size. What remains on the primary side is repaying an Aave borrow position, which is only an exit for someone who minted.

Secondary — intact. The GHO/crvUSD pool still exists and still provides DEX depth. ⚠️ The PegKeeper’s zeroed ceiling ended its role as a defence mechanism, not the pool’s role as liquidity, and conflating the two would understate the exit.

⚠️ So the shape is: a working secondary market and a closed primary one. For a holder who did not mint, a DEX is now the exit, and the fixed-price backstop that would have bounded its worst case is not there.

4 · Dependencies — 6.0

⚠️ GHO’s dependencies concentrate in Aave itself, and that is not diversification — it is one mature counterparty. The lending pool, the collateral basket and all eight facilitators are Aave-governed. A governance failure or a protocol-level incident reaches every leg at once.

⚠️ Docked specifically for concentration inside that: 44.3% of supply sits in a single facilitator, above the 25% single-name line this coverage applies elsewhere.

⚠️ The seized GSMs and the zeroed PegKeeper are scored under axis 1 and are deliberately NOT re-punished here. Same failure, counted once. A reader comparing this axis against the peg axis should not read the 6.0 as disagreement — it is the absence of a double-count.

What the concentration is not: the counterparty is a long-lived, heavily audited protocol with public governance, which is why this axis sits at 6.0 rather than lower. The finding is that there is one of it.

5 · Contract & Admin — 6.0

⚠️ The largest facilitator is not what its label says.

It holds 310,000,000 GHO — 44.3% of all supply — and is labelled “GhoDirectFacilitator GSMs Mainnet”. That name implies a GHO Stability Module: fiat-stable collateral swapped at a fixed price, the arrangement that makes a large minting bucket comfortable to reason about. Its bytecode implements no such thing. Selectors enumerated from the deployed contract at 0xE9ac5231…27d2:

PRESENT   GHO_TOKEN() · mint(address,uint256) · burn(uint256) · AccessControl roles
ABSENT    GSM() · gsm() · getGsmList() · UNDERLYING_ASSET() · POOL() · VAULT()
ABSENT    getRoleMember() · getRoleMemberCount()

Every GSM accessor is missing, and ⚠️ the last line matters most: without getRoleMember() or getRoleMemberCount(), the addresses holding the mint role cannot be listed from chain state at all. So: 44.3% of GHO is minted by a role-gated contract whose authorised callers cannot be enumerated on-chain.

What targeted checks did settle

⚠️ Enumeration reverts, but hasRole answers. Measured 2026-08-30:

hasRole(DEFAULT_ADMIN_ROLE, X)      310M facilitator     GHO token
  Aave Governance v3 Executor L1          TRUE               TRUE
  0x3765a685… (the deployer EOA)          false              false
  GHO token (negative control)            false                —

✅ The deployer EOA holds DEFAULT_ADMIN on neither. ✅ The Aave DAO executor holds it on both, so authority over the token and over the 44.3% facilitator sits with governance rather than a bare key.

⚠️ Three bounds, and this is a narrowing rather than a clearance:

  1. ⚠️ “The executor holds it” is not “only the executor holds it.” getRoleMemberCount reverts on both contracts, so the holder set is confirmed non-empty and cannot be closed.
  2. ⚠️ DEFAULT_ADMIN grants the mint role — it is not the mint role. Who currently holds mint was not tested.
  3. ⚠️ The token is not at the standard EIP-1967 slots — impl and admin both read 0x0 on 8,046 bytes of code. That is not immutability. The upgrade path is simply not where it is normally found, and has not been located.

The cap is the mitigation, and omitting it would overstate the finding

⚠️ The bucket is at its ceiling: 310,000,000 minted against a 310,000,000 cap. Zero headroom. It cannot mint one more GHO without a governance action raising capacity. That is a real constraint, publicly visible, and it bounds the exposure the authority gap creates. A page stating the enumeration problem without stating the cap would describe a larger risk than exists.

Still unestablished, and that is different from unknowable: where the 310M was minted to, not traced; and who holds the mint role, not enumerable through the standard accessors. Both have resolution paths — role-grant events and mint destinations are traceable with the right log queries.

Re-score trigger, written to be checkable by anyone rather than a matter of judgement: any governance action raising the 0xE9ac5231…27d2 bucket above 310M. That would remove the constraint currently bounding the authority gap.

6 · Issuer — 6.5

Aave DAO — one of the longest-running and most heavily audited protocols in the category, with public governance, on-chain voting and a mature executor structure. Authority over both the token and the largest facilitator sits with the governance executor, verified above rather than assumed.

⚠️ What holds this axis at 6.5 rather than higher is disclosure rather than conduct. The mint-role holders are not published anywhere and are not readable from chain; a facilitator carries a name that does not describe what it implements; and the upgrade path on the token is not at the standard slots and has not been located. None of that is misconduct. All of it means a holder must take on trust things that could be made checkable.

Who should avoid this

  • Anyone who needs to know who can mint. For 44.3% of supply, that list is not readable from chain — narrowed by the hasRole checks above, but not closed (axis 5).
  • Anyone treating the “GSM” label as a description of backing. It is a name, not an implementation (axis 5).
  • Anyone relying on a fixed-price exit. Both stability modules are seized; a DEX is the exit (axes 1 and 3).
  • Anyone sizing on growth. Supply has not moved since 2026-08-13 and most of it is at ceilings (axis 2).

What to watch

  • ⚠️ Any capacity raise on the 310M bucket. This is the re-score trigger and the single most consequential thing that could change here (axis 5).
  • Whether either GSM is unseized, which would restore the par-defence route and the fixed-price exit together (axes 1 and 3).
  • Whether the mint role is ever made enumerable, by a facilitator upgrade or by governance publishing the holders (axis 5).
  • Whether supply moves at all. Exactly zero since 2026-08-13 is unusual enough that a resumption is itself information (axis 2).
  • Mint destinations for the 310M, if traced — the open question with the clearest resolution path (axis 5).

Revision history: 2026-08-29 (third) — GSM findings re-verified and the sourcing sharpened; no score change. Both Stability Modules re-read on 2026-08-29 and unchanged across sixteen days — getIsSeized() true, getIsFrozen() false, zero GHO, exposure caps 0 — with the GHO PegKeeper at 0x53876B15…C50b confirming debt() of 0.00. ⚠️ Stated as “verified 2026-08-13, unchanged 2026-08-29” rather than simply re-dated to today, because persistence is the more informative claim here. A module seized today might be mid-incident; one that has held seized, unfrozen, empty and capped across sixteen days is behaving like a decommissioned component, which turns the wind-down reading from an inference into an observation. ⚠️ Each module’s identity is now established by reading UNDERLYING_ASSET() rather than by trusting its name — the same discipline this report applies to the facilitator, and for the same reason: on this asset the labels have already proven the least reliable thing on the contract. The Curve governance action remains reported rather than read. 2026-08-29 (second) — ⚠️ A driver of these scores was missing from the page: both mainnet GHO Stability Modules are seized, and Curve’s GHO PegKeeper debt ceiling is 0. No score change — the marks already reflected this; the page simply did not say so, which is worse than it sounds, because nothing about it looked incomplete. The facilitator finding above concerns issuance authority; this one concerns whether the peg has a working defence, and that is the more retail-relevant half. seized = true with frozen = false is a wind-down rather than a pause — freezing is what you do to a module you intend to restart. Both hold zero GHO and zero underlying with exposure caps of 0, so no route exists to arbitrage sub-par GHO back toward $1, which is what makes Curve’s decision legible rather than assumed. ⚠️ Dated, not current: the seizure states and the zero ceiling are on-chain reads taken 2026-08-13 and not re-read since. The Curve governance action connecting them is reported rather than read — gov.curve.finance returns 403 from two environments. Sourced from this site’s own crvUSD report, which already carried the finding with full contract addresses. ⚠️ The internal source could not re-verify it because its own record stored those addresses truncated, and a reconstructed address that answers is not evidence about the contract you meant — so the addresses were recovered from the published page rather than guessed. 2026-08-29 — first publication of GHO on this site, staged pending a publication decision. ⚠️ The central finding corrects this coverage’s own prior framing: the largest facilitator, holding 310,000,000 GHO or 44.3% of supply, is labelled “GhoDirectFacilitator GSMs Mainnet” and implements no GSM machinery at all — every GSM accessor is absent from its bytecode, enumerated rather than inferred. Previous descriptions of GSM-minted GHO as fiat-stable-backed at a fixed price describe an architecture this contract does not have. ⚠️ getRoleMember() and getRoleMemberCount() are also absent, so the addresses authorised to mint cannot be listed from chain state — the accurate statement is that 44.3% of GHO is minted by a role-gated contract whose callers are not enumerable. The cap is carried with equal weight because omitting it would overstate the finding: the bucket is at 310M of 310M with zero headroom and cannot mint further without a governance capacity raise. Two items are recorded as not established rather than unknowable — where the 310M was minted to, and who holds the role — and both have resolution paths; the deploying address is a zero-byte EOA, but deployment does not establish role-holding and no such claim is made. Supply is frozen: eight facilitator levels sum to exactly 699,000,000.00 against the same totalSupply, every level identical to the 2026-08-13 read, against +7.7% growth in the preceding week — 0.0% over 16 days with 65.8% of supply at hard caps. Scores held at 5.5 because the new facts point both ways and neither is measured well enough to move a number; re-score trigger is any governance action raising that bucket above 310M.