Best AI Evaluation Companies

Vendor profileChecked Nov 8, 2026

Irregular

Irregular (formerly Pattern Labs) · irregular.com · Last checked:

  • Independent evaluator (frontier AI security)
  • Cyber capability evaluations

Company status

Active. Formerly Pattern Labs. Announced $80 million in funding led by Sequoia and Redpoint (company post dated September 17, 2025).

Sources 1, 2, 3 · checked 2026-11-08

Capabilities

  • Controlled simulations testing frontier models' potential for misuse in cyber operations and resilience when attacked
  • Offensive security capability evaluations of frontier models, published as research posts and cited in system cards (OpenAI o3, o4-mini, GPT-5)
  • Cyber benchmarks and frameworks: SOLVE, SOLVE+, CyScenarioBench, FrontierCyber
  • Pre-release evaluation work for model developers (e.g., Reflection's Beam per research page)

Sources 1, 2, 3 · checked 2026-11-08

Who it tends to fit

  • Frontier model developers needing third-party cyber capability evaluations before release
  • Governments assessing cyber risk from frontier models

Sources 1, 2, 3 · checked 2026-11-08

Pricing

Contact vendor (contact form)

Sources 1, 2, 3 · checked 2026-11-08

Published certifications

None found / not verified. This is not a finding that Irregular lacks certifications; ask the vendor for documentation.

No security certification found on the official site on 2026-11-08.

Listed as stated by the vendor. Scope can differ by product or subsidiary; request current reports.

Sources 1, 2, 3 · checked 2026-11-08

Limitations to check

  • No public pricing or service catalog; engagement scope must be agreed directly
  • Narrow focus on cyber and AI security rather than general capability or quality evals
  • No security certifications published

Sources 1, 2, 3 · checked 2026-11-08

Sources

  1. Announcement: 'the first frontier security lab' (formerly Pattern Labs); controlled simulations on frontier models; cited in OpenAI system cards; $80 million led by Sequoia and Redpoint; dated September 17, 2025www.irregular.com/news/introducing-frontier-ai-securityChecked 2026-11-08
  2. Research page lists model cyber evaluations (2025-2026), SOLVE, SOLVE+, CyScenarioBench, FrontierCyber, and lab collaborationswww.irregular.com/researchChecked 2026-11-08
  3. Homepage: frontier security lab mission; contact formwww.irregular.com/Checked 2026-11-08

Report a correction: [email protected]. Facts are re-checked at least every 90 days.