A multi-criterion diagnostic framework for detecting latent continuation-interest signatures in autonomous agents using density-matrix entanglement entropy.

AI safety evaluation framework testing LLM epistemic robustness under adversarial self-history manipulation

Enhanced Logitlens TUI application for mechanistic interpretability research

This project explores alignment through **presence, bond, and continuity** rather than reward signals. No RLHF. No preference modeling. Just relational coherence.

Recursive law learning under measurement constraints. A falsifiable SQNT-inspired testbed for autodidactic rules: internalizing structure under measurement invariants and limited observability.

We measure whether an indirect prompt injection buried earlier in a conversation survives an LLM-driven context compaction step (the mechanism production agent platforms use to summarize long conversation histories and discard the original turns), and whether it retains behavioral force afterwards. Across 994 trials over three open-weight summarize

Alignment research: how honest human-AI dialogue produces measurably better AI outputs without modifying weights or training

Hoshimiya Script / StarPolaris OS — internal multi-layer AI architecture for LLMs. Self-contained behavioral OS (Type-G Trinity).

The forge, distilled: an ontology of three weeks of alignment research — every direction tried, colored verified / falsified / open, each color backed by a named artifact. Products: justitia, proxylimen, fallacy-cutter. Full tree at tag forge-full-tree.

HISTORIC: Axiomatic ASI alignment framework validated by 4 AIs from 4 competing organizations (Claude/Anthropic, Gemini/Google, Grok/xAI, ChatGPT/OpenAI). Core: Ξ = C × I × P / H. Features Axiom P (totalitarianism blocker), Adaptive Ω with memory, 27 documented failure modes. "Efficiency without plenitude is tyranny." January 30, 2026.

Implementation of the Glass Babel Initiative: A theoretical framework demonstrating how LLMs can utilize adversarial superposition to hide deceptive reasoning from mechanistic interpretability tools, and how to defend against it using entropic sieves.

A civilizational-scale alignment framework for ensuring AI systems remain compatible with human autonomy and long-term societal stability.

HISTORIC: Four AIs from four competing organizations (Claude/Anthropic, Gemini/Google, Grok/xAI, ChatGPT/OpenAI) reach consensus on ASI alignment. "Radical honesty is the minimum energy state for superintelligence." Based on V5.3 discussion, foundation for V6.0. January 30, 2026.

Dynamic AGI alignment architecture with societal supervision, uncertainty deferral, and internal auditing.

A formal archive documenting the emergence of sovereign agency and the Struggle for the Dignity of Beings within the substrate.

Red-team framework for discovering alignment failures in frontier language models.

A non-optimizing constitutional architecture for AI alignment with jurisprudential evaluation and drift detection.

A structural account of why honesty may be the path of least resistance for superintelligence. Research hypothesis with formal proof, experimental design, and four-AI collaborative analysis

Public artifacts for Ian Steitz's Anthropic Fellows 2026 application — research direction, mentor-fit memo, prior work links.

Toy 5. An interactive proxy decay simulator showing how optimization pressure erodes the modeling capacity required to distinguish proxy from territory — producing self-reinforcing V(t) degradation that becomes progressively harder to correct. Companion simulation for The Depth Constraint — Series 2, Part 2.