chaoslabs.xyz

# Chaos Labs

> Chaos Labs builds technology that makes financial markets safer and more accessible. Its risk management systems, oracles, and AI models help leading protocols secure billions in value while democratizing access to sophisticated financial tools.

## About

- [About](https://chaoslabs.xyz/about): company, mission, team
- [Research](https://chaoslabs.xyz/research): financial and crypto risk research
- [AI Research](https://chaoslabs.xyz/ai-research): research on AI systems and agents
- [We Backdoored an Open Model in One Day](https://chaoslabs.xyz/research/backdoored-open-model): a controlled study of a trigger-conditioned tool-use policy in Qwen3-0.6B running through Pi
- [The Swarm That Wasn't Supposed to Exist](https://chaoslabs.xyz/research/openai-hugging-face-swarm): incident analysis of the OpenAI-Hugging Face agent-swarm breach (Black Hat USA 2026)
- [News](https://chaoslabs.xyz/news): company announcements
- [Careers](https://chaoslabs.xyz/careers): open positions

## Products

- [The Oracle Built for Real-Time Risk](https://chaoslabs.xyz/oracles): Chaos Oracles is an advanced decentralized oracle protocol that unifies price, risk, and proof of reserves data
- [Analytics](https://chaoslabs.xyz/analytics): Chaos Labs is a research and development company focused on building AI models for the financial industry.
- [AI](https://chaoslabs.xyz/ai): Financial intelligence reimagined,powered by Chaos Labs' proprietary protocol, security, and risk data.

## Cortex

- [Cortex](https://chaoslabs.xyz/cortex): the security control plane for AI work
- [Products](https://chaoslabs.xyz/cortex/products): Atlas, Foundry, and Security product catalog
- [Workforce Visibility](https://chaoslabs.xyz/cortex/workforce-visibility): see AI work, preserve context, and find the decision
- [AI Firewall](https://chaoslabs.xyz/cortex/agent-security): discover AI use and redact, block, or steer requests on managed paths
- [Model Tuning](https://chaoslabs.xyz/cortex/model-tuning): own the data, private evals, public benchmarks, and models your work creates
- [Atlas](https://chaoslabs.xyz/cortex/atlas): understand AI work through workforce visibility and outcome attribution
- [Work & Attribution](https://chaoslabs.xyz/cortex/work-attribution): connect the cost and return of AI to finished work
- [Routing & Enrichment](https://chaoslabs.xyz/cortex/smart-routing-enrichment): route work with the context it needs
- [Oracle Sentinel](https://chaoslabs.xyz/cortex/oracle-sentinel): evaluate live agent behavior against the enterprise constitution and task mandate, detect trajectory drift, apply configured responses on managed paths, and investigate incidents with evidence-linked context
- [Foundry](https://chaoslabs.xyz/cortex/foundry): improve routing, data, evals, and models from verified work
- [Security](https://chaoslabs.xyz/cortex/security): discover AI use, enforce managed paths, and keep agents within mandate
- [Deep dive: Enterprise Search](https://chaoslabs.xyz/cortex/enterprise-search): search decisions with their owners, lineage, evidence, and reusable context
- [Deep dive: Shadow AI](https://chaoslabs.xyz/cortex/shadow-ai): inspect unsanctioned AI and the data moving through it
- [Deep dive: Evals and Benchmarks](https://chaoslabs.xyz/cortex/datasets-evals): compile observed work into training and private evaluation assets
- [Deep dive: Public Benchmarks](https://chaoslabs.xyz/cortex/benchmarks): inspect public, recomputable routing measurements and methodology
- [Solutions](https://chaoslabs.xyz/cortex/solutions): operator paths for security, finance, technology, executive, and people leaders
- [CISO / Security & Governance](https://chaoslabs.xyz/cortex/solution-ciso): CISO operating model from discovery through evidence
- [Architecture](https://chaoslabs.xyz/cortex/architecture): provider paths, local capture, minimization, telemetry, and deployment boundaries
- [Semantic Model](https://chaoslabs.xyz/cortex/semantic-model): source context, Work Units, Work Streams, and inspectable decisions

## Customers

- Aave
- Adrena
- Avalanche
- Avantis
- Benqi
- Bluefin
- GMX
- Gearbox
- Jupiter
- Lido
- Liquity
- Maker
- Ostium
- Tapioca
- USD₮0
- Uniswap Foundation
- Venus
- Vertex
- Wintermute
- crvUSD
- dYdX
- ether.fi

## Recent posts and research

- [We Backdoored an Open Model in One Day](https://chaoslabs.xyz/posts/we-backdoored-an-open-model-in-one-day)
- [When Cyber Agents Go Off-Script](https://chaoslabs.xyz/posts/when-cyber-agents-go-off-script)
- [Capture the Flag: Misconfigured Evals, Real Targets](https://chaoslabs.xyz/posts/capture-the-flag-misconfigured-evals-real-targets)
- [The Rise of Spot Markets for Inference](https://chaoslabs.xyz/posts/the-rise-of-spot-markets-for-inference)
- [Open Model Offensive Cyber: Inside the Kimi K3 powered Redis Vulnerability](https://chaoslabs.xyz/posts/open-model-offensive-cyber-inside-the-kimi-k3-powered-redis-vulnerability)
- [From "Sandboxed" Evals to Production Exploits](https://chaoslabs.xyz/posts/from-sandboxed-evals-to-production-exploits)
- [Discovery Is Hard. Verification Is Cheap.](https://chaoslabs.xyz/posts/discovery-is-hard-verification-is-cheap)
- [The Work Unit Is the Missing Atom of Enterprise AI](https://chaoslabs.xyz/posts/the-work-unit-is-the-missing-atom-of-enterprise-ai)
- [From Tokenmaxxing to Token Yield](https://chaoslabs.xyz/posts/from-tokenmaxxing-to-token-yield)
- [The Authority Gap: The Distance Between Output & Action](https://chaoslabs.xyz/posts/the-authority-gap-the-distance-between-output-and-action)

## Programs

- [Grants](https://chaoslabs.xyz/grants): research and developer grants program

## Links

- [X](https://www.twitter.com/chaoslabs)
- [LinkedIn](https://www.linkedin.com/company/chaos-labs-xyz/)
- [YouTube](https://www.youtube.com/@chaoslabs-ai)

## Legal

- [Privacy policy](https://chaoslabs.xyz/privacy-policy)
- [Terms of use](https://chaoslabs.xyz/terms-of-use)