Insights

Reports, policy briefs, and analysis shaping AI governance

BLOG POSTMARCH 2026

Why Static Benchmarks Fail Agentic Systems

Static benchmarks were designed for stateless models. Agentic systems operate across time, accumulate context, and interact with environments — none of which point-in-time evaluations can capture.

REPORTMARCH 2026

State of Agent Safety Q1 2026

StayInformed

Subscribe to our newsletter for the latest research findings, evaluation results, and policy analysis — or reach out to discuss collaboration opportunities.

Insights

Why Static Benchmarks Fail Agentic Systems

State of Agent Safety Q1 2026

Archive

Why Static Benchmarks Fail Agentic Systems

State of Agent Safety Q1 2026

AAAI 2026: Agent Safety Dominated the Conversation

The Case for Continuous Safety Monitoring in Production

Regulatory Implications of Autonomous Agent Deployment

Red-Teaming Agentic Systems: Methodology Report

Governance Frameworks for Autonomous AI Research Assistants

NeurIPS 2025: Key Takeaways on Alignment & Red-Teaming

StayInformed