Anand Kannappan
I'm the co-founder and CEO of Patronus AI, a frontier data research lab steering progress toward human-aligned AGI. We develop world models to scale agent simulation data, which researchers use in model training.
Our mission is to enable scalable oversight - which I believe will be humanity's last research problem to solve. We are a series B company backed by Lightspeed Venture Partners, Notable Capital, Stanford University, and AI leaders across the labs.
Previously, I worked on AI at Meta, where I led early efforts around interpretability and causal inference. Before that, I ran a quant hedge fund backed by Mark Cuban and Howard Morgan.
Contact
Writings
- Zeno’s Paradox in AI
September 2026
Research
- FinanceBench: A New Benchmark for Financial Question Answering
- SimpleSafetyTests: A Test Suite for Identifying Critical Safety Risks in Large Language Models
- TRAIL: Trace Reasoning and Agentic Issue Localization
- Lynx: An Open Source Hallucination Evaluation Model
- GLIDER: Grading LLM Interactions and Decisions using Explainable Ranking
- MEMTRACK: Evaluating Long-Term Memory and State Tracking in Multi-Platform Dynamic Agent Environments
NeurIPS 2025 SEA Workshop - Benchmarking Reward Hack Detection in Code Environments via Contrastive Analysis
ICML 2026 - Browsing Lost Unformed Recollections: A Benchmark for Tip-of-the-Tongue Search and Reasoning (BLUR)
ACL 2025 - FigmaTrace: Capturing Creative Nuances in Human Figma Design Workflows
- Defenses & Enablers For Skill Injection Attacks on Terminal Based Agents
- DETOUR: An Interactive Benchmark for Dual-Agent Search and Reasoning
- SpeedrunBench: Challenging LLM Agents on Speedrunning Games