Aave tagged v3.7.0 on 5 August 2026. The changelog reads like maintenance work: a few deleted functions, a pool revision bumped from 10 to 11. Keep reading and something larger shows up. Isolation mode and siloed borrowing, two risk controls that shaped how Aave onboarded volatile assets for three years, have been taken out of the protocol entirely.
What isolation mode actually did
Isolation mode existed to answer an awkward question. A new token wants to be accepted as collateral. Nobody yet knows how it behaves in a sharp drawdown. Listing it normally exposes every other asset in the pool to that uncertainty.
Aave's answer was to ring-fence it. An asset listed in isolation
could be posted as collateral, but a borrower using it could
only take out stablecoins the DAO had explicitly approved, up to
a debt ceiling denominated in dollars, and could not combine
that asset with any other collateral in the same position. The
ceiling was tracked on-chain in a field called
isolationModeTotalDebt.
Siloed borrowing worked from the opposite direction. A reserve carrying the siloed flag meant that anyone borrowing that asset could hold no other borrow position in the pool at the same time.
What the release actually deletes
| Removed | Effect |
|---|---|
setDebtCeiling(), setBorrowableInIsolation() |
Isolation mode can no longer be configured for any reserve. |
setSiloedBorrowing() |
Siloed borrowing is gone from the configurator. |
resetIsolationModeTotalDebt() |
isolationModeTotalDebt is deprecated and now returns 0 in view functions. |
dropReserve() |
Reserves can no longer be removed at the pool level. |
priceOracleSentinel |
The L2 sequencer check no longer gates borrows or liquidations. |
liquidationCall also stops updating isolation debt
counters, and PoolInstance moves its
POOL_REVISION from 10 to 11. Integrators reading
those fields should expect zeros rather than errors.
Why they went
BGD Labs, the team that maintains Aave's core contracts, took the proposal to governance on 16 February 2026. The reasoning was unusually direct for a protocol upgrade: there is "no current need for those modes." Siloed borrowing had never been switched on in production. Isolated collateral saw thin deployment. The stated objective was "v3 codebase complexity reduction," described as a change that "simplifies in a very non-invasive way multiple components of the protocol."
Less code means less to audit and fewer interactions between features that risk teams have to reason about before approving a listing. That second point matters more than it sounds. Isolation mode, siloed borrowing and eMode all constrained what a position could hold at the same time, in overlapping ways.
Isolated eMode picks up the slack
The replacement arrives in the same release. Aave v3.7 adds
configureEModeCategoryIsolated(uint8 id, bool isolated)
and a matching
getIsEModeCategoryIsolated(uint8 id) getter, plus an
extra bool isolated parameter on
setEModeCategory.
Flagging an eMode category as isolated tells the protocol to treat any asset outside that category's collateral bitmap as zero-LTV automatically. Risk teams previously achieved the same separation by setting LTV to zero on assets one at a time, a manual process that had to be repeated for every new listing. The behaviour survives. The bookkeeping around it does not.
If you hold a position on Aave
- Check whether any collateral you posted was listed in isolation mode, since the borrowing restrictions attached to it no longer apply the same way.
- Do not rely on
isolationModeTotalDebtin any script or dashboard. It returns 0. - On layer 2s, borrows and liquidations are no longer paused when the sequencer is flagged as down.
The oracle change deserves its own line
Removing the price oracle sentinel is the piece most likely to surprise people. That component watched layer 2 sequencer health and froze certain actions when the sequencer looked unavailable. It was built as a safety measure. BGD's case for deleting it was that it misfired often enough to do net harm. False positives could, in their words, "block liquidations" and "prevent normal user activity" during periods when nothing was actually wrong. A safety mechanism that halts liquidations during a calm market is not protecting anyone.
What to make of it
This release is a quiet admission that Aave's 2022-era risk framework was over-engineered. Isolation mode and siloed borrowing were designed for a period when the protocol expected to list a long tail of unproven tokens and wanted per-asset guardrails for each one. That listing pattern never really materialised at the scale anticipated, and the machinery built for it sat mostly idle, adding audit surface and configuration complexity for years.
Deleting unused features from a contract holding billions in deposits takes more discipline than shipping new ones, and Aave deserves credit for doing it in a versioned release with per-change documentation rather than folding it into something louder. The trade-off is real, though. Governance has given up two tools it may want back if the protocol ever returns to listing genuinely experimental collateral, and rebuilding a removed primitive under time pressure is harder than keeping a dormant one.
For most borrowers nothing breaks today. For anyone running liquidation bots, dashboards or integrations against Aave's pool interface, the revision bump to 11 and the deprecated fields are worth reading before the next upgrade lands.