Skip to content
SCB 10X Summit

Safeguarding DeFi: Managing Risk in a Decentralized Economy

Omer Goldberg, Melody He

0:000:00

Key takeaways

  • Holistic Risk Engine as Core IP: Chaos Labs built its reputation on an AI-powered risk platform that already parameterizes lending, derivatives, and staking markets for Aave, Jupiter, GMX, dYdX, EtherFi, Renzo, LayerZero, and others.
  • Oracle Suite Differentiation: The firm offers three concurrent oracle feeds—Price, Risk, and Proof-of-Reserves—linking real-time market data to on-chain parameters and eliminating the manual “engineer-on-call” model that once hampered protocol safety.
  • Business-Logic Layer Prevents Exploits: By embedding liquidity depth, price-impact thresholds, and reserve data directly into oracle logic, Chaos Labs blocks common manipulation vectors that have cost DeFi users hundreds of millions of dollars.
  • AFI (Artificial Financial Intelligence) Expansion: Leveraging a proprietary cross-chain data lake and an ensemble of 15-plus specialized models, Chaos AI converts unstructured text, order-book, and on-chain signals into actionable recommendations—extending the firm’s risk edge into a full consumer-facing product.
  • TradFi and Tokenization Ready: The same abstractions used for on-chain assets can ingest tokenized treasuries, money-market funds, or equities, positioning the platform for the next wave of real-world-asset growth.
  • Pragmatic Decentralization Roadmap: Product-market fit and user safety outweigh early decentralization; progressive rollout of decentralized compute and inference layers will follow once performance and SLA targets are met.
  • Massive Early Demand: Chaos AI’s wait-list tops one million users; phased onboarding is under way to manage inference costs that could otherwise reach up to a billion dollars annually.

On Chaos Labs’ Product Architecture

Omer Goldberg explained that what appears to be three distinct products—Chaos AI, Edge Oracles, and Chaos Analytics—are in fact layers of one stack. The base is a Monte-Carlo, agent-based risk engine that streams loan-to-value ratios, liquidation thresholds, and interest-rate curves directly on chain for blue-chip protocols. Above it sits the Oracle network, which automates those updates. Edge then exposes a consumer interface, Chaos AI, that lets users and protocols interrogate the underlying models in natural language. This vertical integration removes human bottlenecks while preserving a single source of truth for all market parameters.

On Oracle Risk and Market Manipulation

Both speakers underscored that “oracle exploits” remain the cheapest path to draining a protocol. Recent examples include a half-million-dollar loss on Avalanche when a thin stable-coin pool spiked for a single block, and the Mango Markets incident in which $200 million was siphoned via an inflated governance token price. Chaos Labs counters this by baking market-depth tests, multi-venue confirmation, and time-weighted thresholds into its feeds. The approach reframes an oracle from a dumb pipe into an opinionated risk firewall.

On Artificial Financial Intelligence

The shift from pure quant signals to LLM-enhanced models began with GPT-3, allowing Chaos Labs to ingest unstructured data such as governance forums, economic news, and developer chatter. Their AFI system orchestrates roughly sixteen specialist models—some optimized for intent extraction, others for math, others for prose—each fed by a proprietary data lake that spans thirty blockchains, centralized exchanges, and macro markets. The result is a context-aware agent that can reason about collateral quality, liquidity fractures, or cross-asset contagion in real time.

On Bridging DeFi and TradFi

Goldberg argued that once data are normalized, AMM swaps, order-book trades, Treasury repos, and tokenized MMFs share common abstractions. Since 10–15 percent of crypto’s market cap is already tokenized real-world assets, any credible risk engine must evaluate both spheres simultaneously. The AFI stack, therefore, treats tokenized T-Bills and ETH collateral as first-class citizens and maintains relational mappings between them.

On Decentralization Strategy

While philosophically aligned with open systems, Chaos Labs prioritizes latency, uptime, and user protection. Goldberg likened the journey to that of Hyperliquid or Sui—initial centralization to reach escape velocity, followed by staged decentralization of compute (e.g., BitTensor subnets), data provenance, and governance. Users care first about three-second responses and loss prevention; ideology follows robustness.

On Go-to-Market and Scaling

Chaos Oracles are already the fastest-growing oracle service, with multiple undisclosed partnerships set to go live in the coming weeks. For Chaos AI, the company is onboarding roughly a hundred users today, monitoring usage patterns, and optimizing inference spend before opening broader access this summer. Melody He flagged that organic wait-list traction plus cross-chain protocol adoption positions the firm as a potential category leader, rivaling incumbents like Chainlink for oracles and Perplexity AI for research.

Key actionable insights

  • Monitor upcoming protocol integrations; each new Chaos Oracle deployment materially reduces systemic risk while expanding the firm’s data moat.
  • Track tokenized asset growth, as AFI’s unified risk modeling may become a de-facto standard for both on-chain and traditional assets.
  • Watch for cost breakthroughs in decentralized inference; viable solutions could catalyze a broader migration from centralized AI APIs to on-chain or hybrid compute models.