4 min read

DeFi Lending has 3 bps Crime Drag on EVM and Solana

Source: http://x.com/i/article/2055785469737091072

Every disruptive financial technology has growing pains. DeFi was no different: early lending markets launched quickly, scaled fast, and absorbed a series of painful exploits as the sector learned how to secure code, collateral, oracles, liquidations and governance in public markets.

History matters but it is not representative of modern DeFi. Historians are also usually poor.

The current observed hack/crime loss rate for EVM and Solana DeFi borrowing and lending, excluding bridge-related incidents, is roughly 3 basis points of lending TVL per year.

The data analysed included hack/exploit tags from DeFi Llama. The key question for hack/crime risk is: how large are realised exploit losses relative to the amount of capital using the market? The probability of 3 in 10000 is approximately equal to the rate of Americans that die by slipping and falling over.

On that basis, DeFi borrowing and lending look pretty good, despite the fear factor.

A Segmented View Of DeFi Hacks

DeFiLlama records

**$7.751bn** of gross losses for "DeFi Protocol" hacks through 2026-05-16, but this covers a broad range of activity.

The all-DeFi figure includes bridges, DEXs, derivatives, gaming-adjacent protocols, wallets, infrastructure failures, and non-lending DeFi.

Bridges are especially important: excluding bridge-related DeFi incidents reduces the gross all-DeFi loss number from

**$7.751bn** to

**$4.518bn**.

Image

Code has as nasty habit of doing precisely what you told it and not what you intended of it.

This segmentation is useful because “DeFi” is not one homogeneous risk pool. A bridge exploit, a DEX oracle manipulation, a wallet phishing attack and a lending-market collateral incident are different risk events.

Lending markets are the most hacked of all protocols and with good reason, they have the most assets sitting idle inside a smart contract.

Image

Lending and AMMs lead the charge because they require assets to be stacked in a contract. In fact almost all of the contract types above bridges are related to either AMMs or Lending markets.

In this article we will focus on lending and borrowing markets.

Loss Rates Have Substantially Improved

DeFi TVL is much larger today than in the early exploit years, and lending markets in particular are more mature, more audited, and more monitored.

For EVM and Solana borrowing and lending, excluding bridge-related incidents, the annualised gross hack/crime loss ratio has fallen sharply.

Image

Euler really knocked it out of the park getting all their assets back. The ensure deviation between gross ands recovered in 2023 is Euler. In fact they actually got back $240m from $197m hack due to FX.

Measured over the trailing 365 days to 16th May 2026, EVM and Solana non-bridge borrowing and lending hack/crime losses were:

  • $30.9m gross
  • $30.1m net after recoveries
  • **$99.6bn**average lending TVL
  • 3.1bps gross loss ratio
  • 3bps net loss ratio

That is roughly 3 basis points of lending TVL per year.

Diversification has Significant Benefits

The DeFi hack distribution is heavily skewed: a small number of mega-events drive a large share of the headline total.

On a log x-axis, hack sizes look closer to a lognormal distribution. That is useful visually because it shows that most incidents are small, while the large losses sit in a thin right tail.

Image

Love a good bell curve

Despite what ChatGPT argues, I think this data makes a pretty good argument that portfolio diversification is a fantastic way to diversify against crime.

From a risk transfer / insurance perspective it's also a pretty good model to justify crime insurance across many protocols with strict limits per carrier.

This is also reinforced by the fact that most hacks tend to be isolated and small in comparison to the lending market TVL.

Moreover, the larger the market, the less the market gets spanked for a large percentage of the assets.

Image

Note there were come hacks that were > 100% of TVL that were added to the 100% bucket. The reason some hacks are > 100% of TVL is because of a few reasons: (1) there is a lag between the TVL reporting date and the hack report date so TVL can change somewhat, or (2) more likely DeFiLlama reports TVL in a way that is different to the actual assets at risk

It's an imperfect measurement but heuristically it is accurate in describing the story which is that majority of hacks are isolated to single component in a lending market, and rarely the entire market, especially in larger markets.

This is extremely important data for insurance and risk transfer.

Recoveries are Non Trivially Important

Recovery rates also improve the realised hack-loss picture.

Across all DeFiLlama DeFi Protocol losses, capped recoveries are about 8% of gross losses.

For EVM and Solana borrowing and lending excluding bridge-related events, capped recoveries are higher, about 20% of gross losses.

Most notably Euler achieved a 100% recovery rate. In fact due to FX fluctuations, I think they actually made $40m !

Recovery rates are particularly high when theft is done from within developed nations with good governance. There is also a possibly lesson buried here around permissioning, but that is for another article.

A Positive Future

DeFi lending hack and crime risk is becoming measurable, segmentable and increasingly low in loss-ratio terms.

The data shows a maturing market where realised exploit losses are small relative to TVL, increasingly observable, and concentrated in identifiable risk categories.

In short, fuck the haters. The maths maths.