WillItSurvive
No grouping · United States

Haize Labs

haizelabs.com ↗

Startup building red-teaming and LLM-judge tools to test AI models before release.

Losing its protection
Assessed 30 July 2026 · Evidence as of 30 July 2026
The structural protection is eroding.
On the recordSince 30 July 2026 · Checkable from H2 2026 · Every call on the record Checkpoint: Whether the July 2026 'Pacing the Frontier' statement leads to a mandate for independent AI audits
The call

Haize Labs sells testing tools to the same AI labs building their own version of that tool, with no rule yet forcing anyone to buy from outside.

Haize Labs helps companies find problems in AI systems before they cause harm. The problem is that Cisco bought a competitor and bundled the same service into its own product, OpenAI and Anthropic are building these tools themselves, and Microsoft released free software that does the core work. The one thing that matters most: whether Haize actually has real, ongoing contracts with OpenAI and Anthropic to do their safety testing, since the company has only claimed this on blogs and newsletters, not in official statements from those labs themselves.

Why
01
AI pressure: OpenAI, Anthropic, Google in-house safety and eval teams. The labs Haize calls customers already run their own safety evals, the exact workflow Haize sells them. Could AI do the core job? Yes, likely.
02
Bigger players: Cisco AI Defense (via Robust Intelligence acquisition, 2024). Cisco bundles red-teaming and AI security into its platform, undercutting Haize on price and distribution. Could a giant copy it? Yes, easily.
03
What protects it. Unverified pilot ties to OpenAI and Anthropic plus early traction on its open-source judge library are the only things keeping it relevant.
What would change this call
Cisco AI Defense or Palo Alto Networks announces a customer win in AI red-teaming/eval bundling. Shows incumbents converting Haize Labs' target accounts via platform bundling.
OpenAI or Anthropic publicly describes an in-house red-team/eval platform replacing third-party vendors. Would confirm the frontier labs are insourcing the exact function Haize Labs claims as its core relationship.
A priced funding round or public statement clarifies whether the $12.5M Dec 2025 seed and $100M valuation are real. Would resolve whether Haize Labs' financial position is stable or under stress.
Key numbers
  • $100M post-money valuation in General Catalyst-led round (2024)
  • About 10 to 20 employees (2026-07-01)
  • Signed 4,679 sq ft NYC office, entire 6th floor (2025-11-17)
Named risks
  • Core red-teaming and judge workflow is commoditized by free open-source PyRIT and native lab tooling. Severity: high
  • Claimed OpenAI and Anthropic relationships are sourced only from vendor-adjacent blogs, not the labs themselves. Severity: high
  • Acquisitive incumbents Cisco and Palo Alto Networks already bought into the same market with bundled offerings. Severity: high
  • Funding timeline is inconsistent, including an unconfirmed aggregator claim of a post-Series A seed round. Severity: medium
  • Small team of roughly 10 to 20 people expanding office footprint without disclosed revenue. Severity: medium
Rivals, as read in July 2026
RivalDirectionEvidence
Cisco (AI Defense, via Robust Intelligence acquisition)losing toCisco bundled red-teaming into its enterprise security suite after 2024 acquisition
Palo Alto Networks (Protect AI)losing toProtect AI folded into existing enterprise security distribution, undated but active by 2026
Microsoft PyRITlosing toFree open-source red-teaming framework actively maintained through 2026-04
Virtue AIcontestingDirect competitor raised $30M Series A in April 2025
CrowdStrikegaining onAdded AI red-teaming services as a security incumbent by 2026-04
Questions people ask
Is Haize Labs profitable?
No revenue figures have been disclosed. Haize Labs is privately held and venture-backed, with a Series A led by General Catalyst valuing it at $100M post-money circa 2024. As of July 2026 no ARR, MRR, or customer revenue numbers have appeared in independent reporting.
Who are Haize Labs' biggest competitors?
Cisco AI Defense (built on its 2024 Robust Intelligence acquisition), Palo Alto Networks via its Protect AI acquisition, and Microsoft's free open-source PyRIT framework all compete directly, alongside funded rivals like Virtue AI, which raised a $30M Series A in April 2025.
Could AI replace Haize Labs?
Haize Labs' core product, automated red-teaming and LLM-as-judge evaluation, is itself built on LLMs. As of April 2026, Microsoft's free open-source PyRIT framework and native safety tooling shipped by OpenAI, Anthropic, and Google replicate much of this workflow at no cost.
Who owns Haize Labs?
Haize Labs is privately held and venture-backed. It was founded in late 2023 or early 2024 by CEO Leonard Tang with Richard Liu and Steve Li, who met as Harvard undergraduates and worked as researchers at UC Berkeley's BAIR Lab, per VentureBeat, June 20, 2024.
How much funding has Haize Labs raised?
A Series A led by General Catalyst valued Haize Labs at $100M post-money circa 2024, with 14 total investors including OVO Fund, Soma Capital, Lux Capital, and Stage 2 Capital, per CB Insights data dated November 17, 2025.
Does Haize Labs actually work with OpenAI and Anthropic?
Vendor-adjacent sources, including a WorkOS blog post from November 6, 2025, and a Substack piece from February 8, 2025, claim Haize Labs red-teamed pre-release Anthropic and OpenAI models. Neither lab has independently confirmed these relationships as of July 2026.
Timeline
  • 2026-07-29Fortune reports 1,200+ AI-lab workers push for government AI slowdown plan, a possible regulatory tailwind
  • 2026-05-20Industry roundup names Haize Labs among six-plus funded red-teaming rivals in crowded market
  • 2026-04-02Report notes Cisco AI Defense, CrowdStrike, and open-source PyRIT all now offer AI red-teaming
  • 2025-12-03Devcuration profile describes Haize Labs actively hiring research, engineering and GTM roles
  • 2025-11-17Haize Labs signs two-year lease for entire sixth floor at 74 Broad Street, NYC
  • 2025-04-18Haize Labs publishes Verdict-Vogent case study on voice-agent evaluation
  • 2025-02-21Haize Labs open-sources Verdict, its LLM-as-judge scaling framework, on GitHub
  • 2025-02-08Substack post claims Haize Labs red-teamed pre-release Anthropic model prototypes
  • 2024-12-08Haize Labs publishes case study on AI21 Labs Jamba model alignment partnership
  • 2024-06-20VentureBeat profiles Haize Labs founders and its jailbreak-testing algorithms
Revision history
30 July 2026
First assessed. Losing its protection. The text above is fixed as of this date and is not edited afterwards.
In the same band
American Airlines GroupSafe Superintelligence Inc.Wayfair
More companies losing its protection →
Sources