Analysis
Published
AgentDojo: When fixing the score changes the test
Fide traces an AgentDojo scoring correction through evaluator selection, older benchmark versions and the records excluded by our own audit.
By Fide AI · 9 min read
Insights by domain
Cybersecurity
Authorized security research, cyber capabilities, system controls, and independent incident investigation.
Published insights
Read →
Essay
Published
Fide’s Approach to Embedded Evaluations
Fide AI’s approach to embedded evaluations combines frontier AI engineering, safety-critical systems research and humanistic and social inquiry to study trustworthy AI.
By Fide AI · 10 min read
Visual investigation
Published
URLQuery: When a web address becomes a program
Fide AI reconstructs a known 29-report URLQuery sequence and audits seven encoded payloads across 187 selected records, tracing submitted instructions, recorded responses and data movement.
By Fide AI · 10 min read
Analysis
Published
DSEWiki: What the records showed, and AI reports missed
Fide AI checked DSEWiki reports against incident records and tracked 78 follow-ups. Higher scores often coexisted with earlier errors or unsupported claims.
By Fide AI · 12 min read
Newsletter
Follow what we find—and what we test next.
Get new findings, the evidence behind them, and the questions guiding our next studies.