Ajna V2 Loses $775K As Oracleless Lending Protocol Gets Drained

Ajna v2 lost roughly $775,400 across seven Ethereum pools this week, and the protocol's own design made this exploit unusual. Ajna runs without any price oracle at all.

The Team's Warning

Ajna's official account posted the warning at 7:58 AM UTC on August 29. “Ajna v2 is aware of an exploit and is investigating unusual movements,” the team wrote. “Please withdraw all funds, repay loans, and refrain from interacting with the protocol.”

Ajna's official X post confirming the exploit, screenshotted Aug 29, 2026.

That blunt instruction is not just caution. Ajna v2 is immutable and has no governance, so a live code patch is not an option. Asking users to exit is the only response tool the design leaves the team.

Attack contracts were deployed around 15:16 UTC on August 28. The first extraction landed roughly an hour later on the cbETH pool, in block 25854888.

The losses split unevenly across seven paired-collateral markets. The syrupUSDC pool absorbed the largest hit at $173,700, followed by wstETH at $159,800. rETH lost $127,400 plus a second extraction of $15,600, and cbETH lost $124,800 plus $12,100 across two separate transactions.

WBTC saw $101,800 walk out, WETH/USDC lost $42,000, and sDAI closed the list at $18,000. Every affected pool paired one collateral type against a single asset. No stablecoin-only or exotic-collateral pool appeared on the loss table.

How Ajna Works

Ajna's pitch has always been oracleless price discovery. Lenders deposit into discrete price buckets. Liquidations then run through internal primitives called kick, take, bucketTake, and settle. Those reference the pool's own Lowest Utilized Price rather than an external feed like Chainlink or Pyth.

There is nothing external to spoof. But the internal accounting has to hold up against manipulation from within a single transaction, and this week it did not.

Inside The cbETH Extraction

On the reconstructed cbETH transaction, the attacker flash-loaned WETH from Balancer. They called bucketTake on a sitting auction, then removeCollateral to pull tokens against freshly minted LP. A follow-on settle call with zero quote repaid let roughly 46.5 cbETH leave the pool on that single step. The Balancer loan got repaid, and the attacker kept the collateral.

Two Theories, No Answer Yet

Independent researchers have not reconciled on a single root cause yet. One theory holds the attacker simply harvested an auction that was already sitting underpriced on the book. A competing theory argues the attacker manufactured the setup in one atomic transaction. That opened a liquidation against a position made to look undervalued at that exact instant.

Security firm Defimon said its detection stack flagged the prepared attack more than an hour before the first exploit transaction landed. A Discord alert sent to the Ajna team, Defimon said, went unactioned. The firm published the full pool-by-pool breakdown several hours after the team's own public acknowledgment.

The Ajna V2 entry on DeFiLlama's hacks tracker, screenshotted Aug 29, 2026.

Not Ajna's First Warning

Older audit literature had already flagged risk in this general neighborhood. Ajna's own audit repository contains prior findings on spurious kicks and LUP manipulation through quote-token removal. Those are categories of risk, not proof this exact 2026 exploit chain repeats any single earlier bug.

Ajna has been here before. In 2023, the team asked users to withdraw funds after spotting a potential exploit path in the pre-v2 version. That earlier warning reads strikingly similar in tone to this week's statement.

DeFiLlama's own hacks tracker classifies the incident as a liquidation logic flaw under the broader protocol logic category. It sits alongside a run of smaller exploits across multiple chains this same week. As of publication, no funds have been returned. No official post-mortem has confirmed which of the two competing theories is correct.