Patronus AI

Type: company (AI-assurance / agent-evaluation vendor) Sector: AI assurance / testing / evaluation & simulation software First seen: 2026-07-01 Last updated: 2026-07-01

Created 2026-07-01 from Patronus AI — $50M Series B & Digital World Models (June 2026) (daily AI-governance vendor-intelligence scan). Single-source so far; capability claims are the vendor’s/investors’ own and not independently verified.

Snapshot

Patronus AI is a San Francisco AI-assurance startup (founded 2023 by former Meta AI researchers Anand Kannappan and Rebecca Qian) that evaluates and stress-tests AI agents before deployment. It sits on Paul’s AI-assurance/evaluation watchlist. It is relevant to the wiki because pre-deployment agent simulation/evaluation is a candidate evidence layer for the testing and robustness obligations regulated firms face as they adopt autonomous agents — even though Patronus is not (on current evidence) positioned as a regulated-FS governance platform [S-2026-06-25-patronus-ai-series-b-digital-world-models].

Positions / Claims they advance

  • Raised a $50M Series B (25 Jun 2026) led by Greenfield Partners (Notable Capital, Lightspeed, Datadog, Samsung participating); ~$70M total raised; reported ~15x revenue growth over the prior year — funding facts corroborated by TechCrunch [S-2026-06-25-patronus-ai-series-b-digital-world-models].
  • Unveiled “Digital World Models” — simulated replicas of websites and internal systems in which agents are stress-tested post-training using reinforcement learning, to surface failures (“hacks”/shortcuts) before deployment; currently offered for software-engineering and finance workflows [S-2026-06-25-patronus-ai-series-b-digital-world-models].
  • Positions its evaluation as running “without any human involvement,” and says it competes mainly against AI labs’ own internal agent-evaluation teams [S-2026-06-25-patronus-ai-series-b-digital-world-models].
  • Vendor-labelled claims (not independently verified): that these are the “first” Digital World Models and that they are “language diffusion world models” [S-2026-06-25-patronus-ai-series-b-digital-world-models].

Relationships

  • relates-to → AI Governance Platforms — supplies a pre-deployment agent simulation/evaluation layer within the agentic-governance sub-theme [S-2026-06-25-patronus-ai-series-b-digital-world-models].
  • relates-to → Model Risk Management and Agentic AI — pre-deployment agent stress-testing is adjacent to validation/challenge of autonomous agents [inference].
  • relates-to → EU AI Act — agent robustness/accuracy testing bears on Art. 15 and human-oversight testing expectations [inference].

Tracked changes

  • 2026-06-25 — Announced $50M Series B (led by Greenfield Partners) and launched Digital World Models for agent training/evaluation; software-engineering and finance use cases [S-2026-06-25-patronus-ai-series-b-digital-world-models].

Open Questions

  • No named EU/UK regulated-FS reference customer; “finance” denotes finance workflows, not confirmed regulated-FS deployments.
  • Whether simulated pre-deployment agent testing produces evidence acceptable as validation under SS1/23, or maps to EU AI Act Art. 15 / DORA scenario-testing thresholds (unverified).
  • Whether an evaluation tool that runs “without human involvement” can supply second-line/independent assurance, or only first-line development evidence.

Sources