theseus: extract claims from 2026-04-06-spar-spring-2026-projects-overview

- Source: inbox/queue/2026-04-06-spar-spring-2026-projects-overview.md
- Domain: ai-alignment
- Claims: 0, Entities: 1
- Enrichments: 3
- Extracted by: pipeline ingest (OpenRouter anthropic/claude-sonnet-4.5)

Pentagon-Agent: Theseus <PIPELINE>

entities/ai-alignment/spar.md

@@ -0,0 +1,41 @@
+# SPAR (Supervised Program for Alignment Research)
+
+**Type:** Research Program  
+**Domain:** AI Alignment  
+**Status:** Active  
+**Website:** https://sparai.org/
+
+## Overview
+
+SPAR (Supervised Program for Alignment Research) connects early-career alignment researchers with expert mentors to work on tractable and important alignment problems. The program's project portfolio serves as a revealed-preference signal about where serious alignment researchers believe the field's most urgent open problems are concentrated.
+
+## Timeline
+
+- **2026-01-01** — Spring 2026 cohort launched with 20+ active projects clustered around verification-defeat mechanisms (observer effect, steganographic CoT, illegible reasoning)
+
+## Spring 2026 Research Portfolio
+
+### Verification and Detection
+- "Pre-Emptive Detection of Agentic Misalignment via Representation Engineering" (neural circuit breaker project): uses representation engineering to detect misalignment vectors (deception, power-seeking) before behavioral execution
+- "Automating Circuit Interpretability with Agents": addresses hours-per-prompt bottleneck in circuit tracing using AI agents
+- "Benchmarking In-Context Intent Inference": evaluating whether models can infer and act on intent without explicit instruction
+
+### Model Organisms
+- "Developing and Evaluating Model Organisms for Misalignment": creating controlled examples of misaligned models for safety research
+- "Building a Model Organism of Illegible Reasoning": targeting o3-style reasoning models with opaque chain-of-thought
+
+### Encoded Reasoning
+- "Encoded Reasoning" project: studying how models use non-transparent encoding in their reasoning traces
+
+### Other Safety Topics
+- "Exploring the safety of continual learning methods for LLM agents": safety under distribution shift
+- "Testing AI Incentives": empirical study of incentive structures frontier models respond to
+- "AIxBio model risk mitigations": biosecurity intersection
+
+## Significance
+
+The clustering of SPAR projects around observer effect, steganography, and illegible reasoning independently confirms B4 mechanisms. Researchers working on solutions are working on exactly the problems that behavioral alignment failures identify.
+
+## Notes
+
+No published results from the representation engineering project as of April 2026. The "model organism of illegible reasoning" project suggests the field has identified illegible reasoning in frontier models as a problem severe enough to require dedicated study infrastructure.