Sign in Subscribe
Web3

The Verifiable Data Problem: Why Trustless Contracts Are Built on Trusted Data

The Verifiable Data Problem: Why Trustless Contracts Are Built on Trusted Data

What Happens When Smart Contracts Trust the Wrong Data?

Smart contracts are marketed as trustless — code that executes exactly as written, verified by consensus, immutable once deployed. But here's the uncomfortable truth: the data feeding those contracts comes from sources that are anything but trustless.

Over $400 million was lost to oracle manipulation attacks in 2024-2025 alone. Harvest Finance bled $34 million in a single transaction. Venus Protocol suffered over $200 million in liquidations across three separate exploits. In April 2025, Loopscale — a lending protocol on Solana — lost $5.8 million because its oracle read an inflated token price that an attacker manipulated on-chain in minutes.

The pattern is identical every time: borrow capital via flash loan, distort an AMM pool's spot price, exploit the protocol that blindly trusts that price, reverse the manipulation, repay the flash loan. All within a single atomic transaction. The contract executes flawlessly. The data was a lie.

Why Can't You Tell Good Data From Bad Data?

This problem extends far beyond DeFi. When a New York lawyer submitted a court filing written by ChatGPT and it turned out to contain entirely fabricated case citations, he was fined $5,000 by a federal judge. The AI didn't malfunction — it generated text that looked correct, and there was no verification layer to catch it. The citations had the right format, the right-sounding names, the right judicial tone. Everything appeared authentic until someone actually checked.

Hospitals face the same crisis. When patient records are breached or altered, there's often no cryptographic proof of what changed and when. AI training datasets are routinely poisoned — a study found that even small amounts of manipulated data can significantly degrade model performance, and the affected models have no way to identify which inputs were corrupted after the fact.

In every domain — legal, healthcare, financial, AI — the core failure is the same: systems consume data without being able to verify its provenance, integrity, or authenticity. And the consequences scale with the stakes.

What Solutions Actually Exist Today?

The standard approach to oracle security relies on two mechanisms. Off-chain oracles like Chainlink aggregate data from multiple independent sources (Binance, Coinbase, Kraken, OKX) through decentralized node operators who submit the median price on-chain. Corrupting this feed requires compromising a majority of independent nodes — difficult but not impossible.

On-chain oracles derive prices directly from AMM pool reserves. A Uniswap pool holding 100 ETH and 200,000 USDC implies a price of $2,000 per ETH. But that price can be changed by anyone with enough capital to move the pool's reserves — and flash loans make that capital essentially free.

Time-Weighted Average Price (TWAP) oracles improve on spot price by averaging across multiple blocks, making sustained manipulation more expensive. But short TWAPs in low-liquidity pools remain vulnerable, and the fundamental issue persists: the protocol still has no way to independently verify whether the data it received is true.

On the decentralized storage side, the landscape is equally fragmented. IPFS provides content addressing — a file's hash is its identity — but offers zero availability guarantees. Your data exists only as long as someone keeps pinning it. Filecoin adds economic incentives for storage, but the developer experience is heavy and verifiability is partial. Arweave guarantees permanent storage through endowments, but the data is immutable and unindexed — fine for archives, unusable for applications that need to update content.

What Does Verifiable Data Actually Require?

True verifiable data needs three properties that existing solutions deliver in isolation, if at all:

Content addressing — data is identified by its cryptographic hash, not its location. You can verify integrity by recomputing the hash. This is IPFS's core contribution, and it works.

Provable availability — you can cryptographically prove that data is actually stored and retrievable, not just that someone claimed to store it. This requires proof-of-storage mechanisms that go beyond Filecoin's sector-level checks.

Programmable access control — data needs rules around who can access it, when, and under what conditions. Seal, launched in September 2025, was the first decentralized on-chain access control system. It uses threshold encryption on Sui's Move object model — data is encrypted, access keys are distributed across validators, and retrieval requires quorum agreement. No single party controls access.

Walrus, the decentralized storage network built on Sui, combines all three through its RedStuff 2D erasure coding system. Data is encoded using a two-dimensional erasure code (think of it as a crossword puzzle where entire rows and columns can be lost without losing any information), split into slivers stored across 100 nodes in 19 countries, and backed by cryptographic Proof of Availability. Each sliver carries a Merkle commitment — you can verify any piece of data against the commitment without downloading the whole file.

Is Anyone Actually Using This?

The adoption numbers suggest this isn't theoretical. Walrus crossed 450TB of stored data with 200+ projects building on it. Team Liquid migrated 250TB of esports data. Allium, a blockchain analytics platform, stores 65TB on the network. Alkimi serves 25 million ad impressions daily through Walrus-stored creative assets. Decrypt moved its entire media library. Grayscale launched a Walrus Trust — the first dedicated investment vehicle for the network's token.

For AI specifically, MemWal provides an SDK for AI agent memory — structured, verifiable storage for agent state, conversation history, and knowledge bases. As AI agents increasingly operate autonomously, the ability to verify what data they've been trained on and what context they're operating from becomes a security requirement, not a nice-to-have.

Why Does This Matter Beyond Crypto?

The verifiable data problem isn't a blockchain problem. It's an infrastructure problem that blockchain happened to make visible. Every system that makes decisions based on external data — AI models, legal databases, healthcare records, financial protocols — challenges closely related to the zero trust security model that modern enterprises are adopting — faces the same fundamental question: how do you know the data is what it claims to be?

As AI agents begin transacting autonomously, as legal systems incorporate AI-generated content, and as healthcare moves to interoperable digital records, the cost of unverifiable data compounds. The lawyer who trusted ChatGPT's citations paid $5,000. The DeFi protocols that trusted manipulated oracles paid hundreds of millions. The next generation of failures will scale accordingly unless verifiable data becomes a foundational layer, not an afterthought.

Sources

Read next

Comments ()