$3.5 Million Against $550,000 of Collateral: How to Check Which Tokens Your DeFi Lending Market Accepts

If you have funds sitting in a decentralised lending market, your risk is not decided by the interest rate but by the list of tokens the protocol accepts as collateral. On September 17, 2026, an attacker on Starknet exploited exactly that list and borrowed around 3.5 million US dollars from the Nostra Finance money market. The collateral was the protocol's own token NSTR, whose entire circulating market capitalisation stood between 550,000 and 590,000 US dollars at the time. The borrowed amount exceeded the market value of the posted collateral by more than five times.

This article explains what is technically new about the incident and what is not — and above all, which four figures you can look up in the documentation of your own lending market before you next put money into it. The check takes about ten minutes and requires no expertise in blockchain programming.

What happened on Starknet on September 17: $3.5 million against $550,000 of market value

Nostra Finance runs a money market on the Ethereum scaling solution Starknet. That is a protocol where users deposit tokens and other users take out loans against posted collateral. The protocol reported the incident on September 17, 2026 at 13:28 UTC through its official channel. A single account had treated an inflated NSTR balance as collateral and then borrowed a basket of far more liquid assets: Ether, STRK, USDC, USDT, WBTC and DAI in the first version.

The security firm PeckShield reported on September 18 at 00:41 UTC that around 1.92 million US dollars had been moved to the Ethereum mainnet, specifically 234.57 Ether and 1.3 million DAI. CertiK arrived at a similar split barely two hours later: roughly 1.55 million US dollars initially remained on Starknet, and around 1.93 million was bridged. The analytics firm SlowMist classified the event as oracle manipulation. Nostra itself has switched off deposits, borrowing, withdrawals and liquidations while the team reconciles the pool balances. How large the final loss turns out to be, and how much of it can be recovered, remains open.

Oracle manipulation explained: how a price feed becomes the attack surface

An oracle is the interface through which a protocol learns what a token is currently worth. Without that information a lending market cannot operate: it has to know how much a posted collateral covers and at what point a position must be liquidated. Oracle manipulation is an attack in which it is not the protocol's program code that is broken, but the price that this code takes at face value.

The distinction matters, because it explains why security audits of the code help little here. The protocol calculates correctly — it merely calculates with a false number. Which price provider was in use in the case of NSTR has to this day not been named in the official statements. Security researchers assume that a trading venue with very thin liquidity was used to push the reported price far above the usual trading level. NSTR was trading between 0.0055 and 0.0059 US dollars at the time of reporting.

Nostra has had this weakness before. On March 24, 2025, the price feeds for xSTRK and sSTRK displayed three times the actual value; xSTRK jumped from 0.1793 to 0.5897 US dollars. A second, independent price provider as a fallback did not exist for those assets.

Why thin liquidity in the collateral token is the real risk

The technical debate usually revolves around oracles. The leverage, however, lies elsewhere. A price can only be moved if the market behind it is thin enough. With a token traded in the hundreds of millions each day, an artificial doubling costs more than the attack brings in. With a token carrying half a million dollars of market capitalisation, a modest stake is enough.

From that follows a rule you can apply without technical tools: the smaller the collateral token relative to what may be borrowed against it, the greater the manipulation risk for everyone else in the same pool. That includes you, even if you have only deposited and never taken out a loan — because in the end the payout came from the shared balance.

Borrow caps and isolation mode

Good protocols limit this risk with two tools. A borrow cap sets how much may be borrowed against a particular collateral in total. An isolation mode locks risky tokens into a separate pot, so that a default there does not spill over into the large pools. Both are stated in the documentation, and both are a number you can look at before you deposit. If the cap is missing entirely, or sits above the market capitalisation of the collateral token, that is a finding and not a detail.

Anyone looking for interest on crypto balances will find different risks at centralised providers, but comparable questions — what collateral sits behind it, and who is liable. A look at our comparison of lending providers shows which models exist in the first place, before you decide between decentralised and centralised.

Checking your lending market's collateral list: these four figures are in every documentation

Every serious protocol publishes its parameters. At Nostra you will find them at docs.nostra.finance; at other protocols the section is usually called Markets, Risk Parameters or Asset Listing. Four figures are enough for a first assessment.

First: which tokens are approved as collateral? Not which ones you can deposit, but which ones can be borrowed against. Those are two different lists, and the second is the dangerous one. Look for the protocol's own token — if it appears as collateral, look more closely.

Second: how high is the borrow cap per collateral? Set it against the market capitalisation of that token. At Nostra, more could be borrowed against NSTR than the entire circulating supply was worth.

Third: does an isolation mode apply? If so, for which tokens, and does your balance sit in the shared or in the isolated pot?

Fourth: who supplies the price? More on that in a moment.

Price source and fallback oracle: how to tell whether a protocol seeks a second opinion

The documentation usually states which price provider is used. Less interesting than the name are the answers to two questions. Does the protocol draw the price from several independent sources, or does everything hang on one? And is there a plausibility limit that freezes a position if a price jumps several fold within minutes?

A fallback oracle is a second price source that steps in or objects when the first delivers implausible values. That is precisely what Nostra lacked in March 2025 for the affected assets. If you find no statement on this in the documentation, that is not an all-clear but an open question you can put to the team in the public channel. A protocol that does not answer has already answered you.

We have asked the same question in earlier incidents. In the Moonwell exploit on Base at the end of August, it was the protocol's own token MAMO whose price was manipulated. The pattern has been repeating for months under changing names.

32 price manipulations in one year: what the numbers and our own count show

The Nostra incident is not an isolated case but the latest point in a series. According to data circulated by the exchange KuCoin in early September, decentralised lending protocols have recorded 32 separate price manipulation attacks so far in 2026 — on that count a record for the sector. Roughly one in eight crypto hacks this year can accordingly be traced back to this form of attack. The largest single loss fell on Tectonic on the Cronos chain, where around 75 million US dollars were borrowed after a price had risen a hundredfold within about 20 minutes. The figure comes from a market participant and not from an independent auditor; it shows an order of magnitude, not an official statistic.

Because we wanted to know how this is reflected in our own coverage, we counted the German-language archive of cryptoticker.io. Method: a query of the editorial database on September 18, 2026 for all German articles since July 1, 2026 whose address contains the terms "exploit" or "hack", followed by a manual review of the hits. Result: 21 documented incidents in eleven weeks. Six of them hit lending markets or their price sources — Ostium in July, Ajna, Tectonic, Moonwell and More Markets at the end of August, plus the compromise of a price service across four chains in early September. This analysis was carried out by cryptoticker.io itself on September 18, 2026.

What we cannot measure with it: how many incidents there were in total. Our count only captures what we reported on ourselves, and articles without the two search terms in the address slip through the net. The number works as an indication of a pattern, not as a damage balance sheet. The clustering is striking all the same: four of the six lending market cases fell within a single week at the end of August.

When a market pauses: why you cannot withdraw when it matters most

The part many underestimate comes after the attack. Nostra stopped not only borrowing but withdrawals as well. From the protocol's point of view that is correct, because otherwise the fastest movers clear out the rest and the damage is distributed unevenly. For you it means: at the moment you most urgently want to get your money out, the door closes.

From that follows an uncomfortable but useful exercise. Ask yourself before depositing how long you could do without this amount if it were frozen for an indefinite period. Not whether you could lose it — that is the obvious question — but whether you will not need it for weeks. Anyone who has parked funds for a tax payment or an upcoming bill does not belong in a money market.

You can also check who is allowed to trigger such a pause. In most protocols that right sits with a small circle of key holders, sometimes with a timelock instance with a lead time. Both are stated in the documentation under Governance or Admin Keys.

Yields that look too good: how to read return against collateral risk

High deposit rates in a money market arise because somebody is willing to pay that rate. Who pays double-digit rates for a loan? As a rule, someone building a position with borrowed money that is meant to return more — or someone who has posted collateral they could not borrow against anywhere else. The second case is the same one that led to the loss at Nostra.

That does not mean high rates are automatically a warning sign. It means the rate tells you where to look. If a conspicuously high rate sits next to an exotic collateral token in the list, you have found the explanation. If only Ether and an established stablecoin are listed, the rate is probably an expression of demand and not of risk.

Centralised or decentralised: which risks you take on in each case

The comparison does not come out clear-cut, and anyone presenting it as clear-cut is simplifying. With a decentralised money market you bear the risk of the parameters, the price feed and the program code. You can read up on all of it, and nobody can withdraw your balance at will — as long as the protocol's key management holds.

With a centralised provider you bear counterparty risk instead: the provider lends your balance on, and whether the underlying collateral holds value is something you generally cannot check. In exchange there is a contractual partner you can address and, at providers with European authorisation, a supervisory authority. Which form suits you depends on whether you would rather read parameters or contracts.

What self-custody changes about this

The third route is not to lend at all. Anyone holding their assets in their own custody forgoes interest and thereby rules out both types of risk. That is not a romantic position but a sober one: a return of a few percent a year does not justify every default risk, and the incidents of recent weeks show that default is not the exception.

Checking a DeFi lending market: what to take away

1. Look at the collateral list before you deposit. Find out which tokens can be borrowed against, how high the borrow cap is per token, and whether it exceeds the market capitalisation of that token. If you want to know in advance which providers even come into question and how interest-bearing models differ, our overview of staking and yield platforms helps with the shortlist.
2. Clarify whether you really need the interest. Work out what a year of yield brings on your amount, and set it against the case where the market pauses for weeks. If the sum comes out close, self-custody is the simpler answer; which devices are suitable is set out in our hardware wallet comparison.
3. Document your deposits and withdrawals as you go. If a protocol enters a wind-down, you will later need records of times and amounts — for your tax return as much as for a possible loss offset. A portfolio tracker takes that off your hands; you will find a selection under crypto tax software and portfolio trackers.

And in case you are already affected: wait for the team's official settlement before responding to offers that promise you a quick recovery. After every major incident, accounts appear that pose as a recovery service and in truth ask for wallet access. The details of the Nostra wind-down are summarised, among others, in the chronology of the security reports.

(As of September 18, 2026. This article is not investment advice. Prices and fee structures change; check the terms with the provider before you buy.)