# Haize Labs

> Haize Labs is a New York-based AI safety and reliability startup that automates the red-teaming, stress-testing, and evaluation of large language models. Founded in 2024 by a trio of Harvard-trained researchers, the company builds algorithms that hunt for the inputs that make AI models misbehave - jailbreaks, failure modes, and edge cases - so they can be fixed before real users find them. Its 'haizing suite' and multi-turn attack engine Cascade are used by frontier model labs including OpenAI and Anthropic, alongside enterprises like Deloitte and MongoDB.

- **Founded:** 2024
- **Headquarters:** New York, New York, United States
- **Founders:** Leonard Tang (Co-Founder & CEO), Steve Li (Co-Founder), Richard Liu (Co-Founder)
- **Team size:** ~19 employees
- **Products:** Haizing Suite, Cascade, Reliability Harness, Model-based Evaluators / Verify, Verdict
- **Notable:** Reached a $100M post-money valuation within roughly a year of founding., Signed both OpenAI and Anthropic as customers - a rare double for a young startup., Built Cascade, an automated multi-turn red-teaming system that outperforms manual jailbreaks on the majority of tested models.

## Products & services

- **Haizing Suite** — A collection of red-teaming, fuzzing, and optimization algorithms that search a model's input space to surface any input that elicits undesired output behavior.
- **Cascade** — An automated multi-turn red-teaming engine that finds conversation trajectories where benign queries escalate into unsafe, jailbroken responses, using tree search and prompt optimization to match expert human red-teamers.
- **Reliability Harness** — Proprietary infrastructure for building and running expert-level AI agents for mission-critical work with measurable reliability.
- **Model-based Evaluators / Verify** — Turns a company's goals for its AI system into automated evaluation rules via synthetic data generation, adversarial attacks, and active learning that sharpens the rules over time.
- **Verdict** — Open-source framework for inference-time scaling of LLMs-as-a-judge, part of Haize Labs' evaluation tooling.
- **j1 reward models** — j1-micro (1.7B) and j1-nano (600M) - very small but capable reward models released open source.

## Achievements

- Reached a $100M post-money valuation within roughly a year of founding.
- Signed both OpenAI and Anthropic as customers - a rare double for a young startup.
- Built Cascade, an automated multi-turn red-teaming system that outperforms manual jailbreaks on the majority of tested models.
- Released open-source tooling including Verdict, j1 reward models, and a public 'get-haized' jailbreak collection.

## Latest updates

- **2024-11** — Released Cascade, an automated multi-turn red-teaming engine, and shared it publicly via the company blog and X.
- **2024-08** — Emerged with a $12.5M seed round led by General Catalyst at a $100M post-money valuation.
- **2025-01** — Named among contributors involved in stress-testing frontier model jailbreak defenses as the AI safety-eval ecosystem matured.

## Links

- Website: https://haizelabs.com
- LinkedIn: https://www.linkedin.com/company/haizelabs
- Twitter/X: https://x.com/haizelabs
- GitHub: https://github.com/haizelabs

---

Profile page: https://yespress.io/haize-labs
Published by YesPress — https://yespress.io
Last updated: 2026-08-05