Evaluation Suitecompany reported
Arena
Adversarial model evaluation for model builders.
Commercial
ArenaPure-play commercialCatalog available · 3
Gray Swan AI
Adversarial red-teaming arenas and safety evals for frontier models
Gray Swan AI, founded by CMU professors Zico Kolter and Matt Fredrikson with Andy Zou, runs adversarial red-teaming arenas, jailbreak competitions, and security evaluations used by frontier labs including OpenAI and Anthropic. Its environments stress-test agent robustness and safety.
Co-founder Zico Kolter chairs OpenAI's safety committee; ran red-teaming arenas for GPT and Claude launches
This legacy directory profile is awaiting claim-level source migration. Existing values are retained, not upgraded to verified facts.
| Website | grayswan.ai |
|---|---|
| Domains | Cybersecurity, Alignment |
| Location | Pittsburgh |
| Team size | 11-25 |
| Founded | 2023 |
| Funding | ~$40M (reported) |
| Founders | Zico Kolter @zicokolter, Matt Fredrikson, Andy Zou |
Capability coverage
Focus areas and technical capabilities are shown separately. Missing technical evidence remains Unknown.
Catalog-evidenced technical capabilities
Unknown. No source-backed catalog capability record is available yet.
Capability catalog
Products and services supported by official company materials. Reviewed 2026-08-16.
Evaluation Suitecompany reported
Adversarial Evaluation
Adversarial evaluation services for AI models.
Commercial
Adversarial EvaluationEvaluation Suitecompany reported
AI Red Teaming
Security red-teaming programs for AI systems.
Commercial
AI Red TeamingEnvironments & datasets
Verified public artifacts and documented private commercial inventory.
No verified catalog items recorded yet. This does not mean the company has no environments or datasets.
Sources
Structured sources have not yet been migrated for this profile.
Improve this record
Company representatives and researchers can propose sourced changes. Submissions do not directly overwrite editorial data.
Similar startups
Also listed under Cybersecurity, Alignment.
| Name | Domains | Type / catalog | Location | Team | Funding | Raising |
|---|---|---|---|---|---|---|
| Rise Data Labs RL environments and tasks across multiple domains, supported by a large expert network | Computer Use, Coding, Finance, Cybersecurity, Legal, Data Labeling, Enterprise, Multi-Domain, RLHF, Custom Environments | Data + environments 3 artifacts | New York, United States | 11-25 | - | Unknown |
| Andon Labs Long-horizon autonomy benchmarks like Vending-Bench | Long Horizon, Alignment | Pure-play commercial Catalog unknown | San Francisco | 1-10 | Seed (Y Combinator) | Unknown |
| ARIMLABS Security and long-horizon environments for agentic AI | Cybersecurity, Long Horizon | Pure-play commercial Catalog unknown | Warsaw | 11-25 | - | Unknown |
| Incalmo Offensive-security environments for AI agents | Cybersecurity | Pure-play commercial Catalog unknown | San Mateo | 11-25 | - | Unknown |
| Invariant Labs Security testing and analysis for AI agents; acquired by Snyk | Cybersecurity, Alignment | Acquired / inactive Catalog unknown | Zurich | 1-10 | Acquired by Snyk (June 2025) | Unknown |
| Quesma Security-domain RL environments and binary analysis evals | Cybersecurity, Coding | Pure-play commercial Catalog unknown | Warsaw | 11-25 | - | Unknown |