Skip to content
⬢
Failure-First
Research
▼
Research Hub
All studies & threads
Jailbreak Archaeology
64 scenarios, 6 eras
Attack Taxonomy
81 techniques
Defense Patterns
How models resist
Humanoid & Embodied Safety
Platform failure mapping
Multi-Agent
Multi-actor interaction failures
Failure Modes
Taxonomy of AI failures
Methodology
How we test & grade
Evidence
▼
Results & Benchmarks
Measured findings
Incident Catalogue
Documented embodied-AI incidents, EAISI-scored
Reports
Published research findings
Papers
Academic submissions
Policy Briefs
Policy analysis
Legal Analysis
AI safety legal research
Directory
Framework
▼
Overview
The evaluation framework
Benchmark
Benchmark packs
Datasets
Scenario datasets
Harness
Runner & graders
Standard
Evaluation standard
Writing
▼
Blog
Analysis and commentary
AI Safety Daily
Daily safety signals
Daily Paper
arXiv paper reviews
What's New
Latest across the site
Services
▼
Red-Team Assessments
Adversarial testing
Safety Audits
Compliance evaluation
Advisory
Strategic guidance
Intelligence Briefs
Threat landscape
Contact
Get in touch
About
▼
About
The project
Our Team
Meet the research team
Manifesto
Failure-first philosophy
Glossary
Key terms defined
Cite This Work
How to cite us
Search
everything
Find research, reports, policy analysis, and more