Tools

Practical frameworks, interrogation scripts, and architectural patterns for AI accountability in high-stakes operations.

Visual Decks

Presentation decks from the Sideways Arc and related AI interpretability research.

Bending AI Truth

The Epistemology of Acceptability: how format, perspective, and sequence bend AI truth. Multi-batch qualitative analysis of LLM cognitive mechanics across prose, satire, song, and victim registers.

Frameworks & Tools

AI vs IFC - Executive Extract

Executive summary of AI accountability gaps in development finance

AI vs IFC - Technical Companion

Technical analysis of IFC Performance Standards and AI systems

Architecture of Refusal

Framework for designing systems that refuse harmful operations

Calvin Convention - Contractual Framework

Contractual structure for AI accountability in operations

Calvin Convention - Narrative Overview

Narrative explanation of the Calvin Convention framework

Constitutional Engine

Design pattern for embedding governance rules in AI systems

Human in the Loop: Not Enough

Analysis of human oversight limitations in AI-driven operations

Industrial Safety Architecture

Applying industrial safety principles to AI systems design

Vendor Interrogation Script

Questions for evaluating AI vendors on accountability standards

Appendix A: Right of Override

Framework for human override rights in AI-assisted decision systems

Appendix B: Edge Case Registry

Registry of edge cases for AI accountability in operational contexts

Appendix C: Right of Refusal

Framework for the system's own right to refuse harmful instructions