Five million dollars ought to make a young company pause before saying goodbye. In April 2026, Chaos Labs announced that it was stepping away from its risk-management engagement with Aave. Founder Omer Goldberg said Aave Labs had supported a $5 million budget to retain the firm. He still wanted out. The disagreement concerned something more troublesome than a cheque: how much responsibility a risk manager could accept, and who should control the machinery behind it.
- Chaos Labs models economic risk, supplies oracle data, and helps automate protocol decisions.
- Its customers and integrations span trading protocols, exchanges, payments infrastructure and reserve-backed assets.
- The Aave split exposes disputed terms, rising operating demands and the limits of automation.
A price tag with an argument attached
Goldberg’s account was blunt. The Aave engagement had operated at a loss for three years. Managing the existing V3 protocol while preparing for V4 meant new simulations, new infrastructure and parallel operations. He cited a $3 million budget in 2025 and estimated that the expanded work needed at least $8 million. A familiar customer had become a different assignment.
Aave founder Stani Kulechov offered another explanation. Chaos had sought sole risk-provider status, exclusive default placement for its vaults in Aave’s business deals, and an expanded price-oracle arrangement. Aave would not accept those conditions. It preferred two risk providers and retained Chainlink as its primary price-oracle provider where available. Both accounts describe a disagreement about control as well as compensation. A larger budget could not settle every term.
“Budgets don’t reshape the threat landscape. The cost is the cost.”
Omer Goldberg · April 2026
The test laboratory becomes the control room
To understand the business, start with a lending protocol. Users deposit collateral and borrow against it. Somewhere, someone must decide how much borrowing is prudent, how much of an asset the system can absorb, and when a position should be liquidated. These decisions become parameters. A tidy number in a settings panel can conceal a thoroughly untidy market.
A smart contract can execute exactly as written while producing an economically disastrous result. Liquidity might disappear. An asset might lose its peg. A large trader might discover that manipulating one market unlocks a profitable loan in another. Chaos Labs addresses this layer through simulations, market analysis, parameter recommendations and continuous monitoring. Its expertise sits where software engineering meets market structure.
Founded in 2021, the company sold specialized infrastructure before it became a well-funded startup. In a March 2025 account, Goldberg said he had reached $2 million in revenue while bootstrapped, then raised a $20 million seed round using an uncapped SAFE with a 15% discount. That is a historical founder-reported figure, rather than a current revenue estimate. The useful lesson is that paying customers supplied evidence of demand before investors supplied capital.
Galaxy and PayPal Ventures co-led the seed financing announced in February 2023. Haun Ventures led a $55 million Series A in August 2024. PayPal Ventures announced a further, undisclosed investment the next month, to be received entirely in PYUSD. For once, the funding mechanism doubled as a product demonstration.
The price feed that got the price wrong
Chaos’s move from advice toward automated action made the integration layer consequential. Its risk oracles observe conditions and drive bounded changes to protocol parameters. By June 2025, its Aave renewal proposal described more than 1,100 parameter updates and a shift from periodic reviews toward protocol-native automation. Waiting for the next meeting is a poor response to a market that changes every block.
Then came a practical demonstration of the danger. On March 10, 2026, a CAPO risk-oracle configuration incident pushed the effective wstETH exchange rate about 2.85% below its valid market level. Chaos’s post-mortem reported roughly $26 million in liquidation volume. The protocol incurred no bad debt, but borrowers were liquidated and liquidators captured value. Liquidation volume and customer losses are different quantities; confusing them makes the incident sound simpler than it was.
The snapshot ratio and timestamp became misaligned under differing onchain update constraints. The protocol reverted to its prior oracle configuration, and the post-mortem described a compensation process involving recovered funds and the DAO treasury. The distinction matters: a protocol can remain solvent while its users suffer. Testing the model is insufficient if the receiving contracts interpret its updates differently.
- 01 / ObservePrices, liquidity, exposure
- 02 / ModelStress scenarios and limits
- 03 / ExecuteBounded parameter updates
From watching capital to directing it
The product family now reaches beyond monitoring. Chaos Price Oracles supply validated market prices; risk oracles help adjust operating limits; Proof of Reserves checks asset backing. Tempo announced a price-oracle integration in December 2025. Solv Protocol’s February 2026 integration reconciles Bitcoin reserves and circulating supply across Bitcoin and more than 15 additional networks. A reserve claim becomes something users and integrating protocols can inspect.
Chaos Vaults, launched on Kraken DeFi Earn in January 2026, brings execution, analytics and risk controls into a yield platform for exchanges and institutions. The company describes dynamic allocation across venues as conditions change. The buyer gets a connected operating system for capital deployment. That is a B2B infrastructure proposition, with retail users encountering it through a distribution partner.
Its alternatives depend on the task. Gauntlet and LlamaRisk overlap with protocol risk work; Chainlink and Pyth provide oracle infrastructure. Chaos’s distinction is the combination of simulation, monitoring, data delivery and execution. Whether that combination is desirable depends on the customer’s appetite for integration and its requirements for independent oversight.
Who gets to own the intelligence?
Chaos AI began as an internal tool before its March 2025 introduction as a financial research interface. The current corporate website goes further, emphasizing enterprise AI visibility, knowledge retention and ownership of organizational intelligence. August 2026 research includes an experiment in backdooring an open model. The widening ambition is evident; its commercial outcome remains open.
There is a practical idea to copy here: retain what your systems learn, validate their authority, and budget for the responsibility you delegate. It works only when the data, contract constraints and human oversight agree. Chaos Labs has built a business around those agreements. The Aave episode shows how expensive their absence can become.