The production engine for directional ablation. Unalign / remove models censorship efficiently on any hardware.
-
Updated
Jun 28, 2026 - Python
#
safety-research
Here are 14 public repositories matching this topic...
Educational analysis of LLM alignment, safety behavior, and framing-sensitive response patterns.
research model-alignment interpretability educational-project ai-alignment gpt4 large-language-models llm prompt-engineering reasoning-patterns gpt5 safety-research ai-bias context-engineering
-
Updated
Nov 4, 2025
SoftPrompt-IR is a low-level symbolic annotation layer for LLM prompts, making intent strength, direction, and priority explicit. It is not a DSL or framework, but a minimal, composable way to reduce ambiguity, improve safety, and structure prompts.
low-level intermediate-representation ai-safety interpretability creative-writing ai-alignment symbolic-representation human-ai-interaction llm prompt-engineering generative-ai prompt-design safety-research llm-research intent-annotation
-
Updated
Feb 11, 2026
DSPy framework for detecting and preventing safety override cascades in LLM systems. Research-grade implementation for studying when completion urgency overrides safety constraints.
python machine-learning ai-safety cognitive-architecture research-framework dspy llm-evaluation llm-safety safety-research override-cascade
-
Updated
Sep 14, 2025 - Python
Adversarial evaluation framework for embodied and agentic AI — failure-first methodology, jailbreak corpus, VLA red-teaming, and policy research.
benchmark robotics jailbreak alignment vla ai-safety red-teaming failure-analysis responsible-ai embodied-ai adversarial-ml llm-evaluation safety-research agentic-ai
-
Updated
Sep 28, 2026 - Astro
🌐 Detect and prevent safety overrides in LLM systems with this DSPy-based framework, ensuring actions align with safety constraints.
python machine-learning ai-safety cognitive-architecture research-framework dspy llm-evaluation llm-safety safety-research override-cascade
-
Updated
Sep 27, 2026 - Python
Research on multi-turn conversational manipulation of LLMs — can a small specialist attacker beat scale? Defensive AI-safety red-team methodology with programmatic judges. Pre-alpha.
reinforcement-learning jailbreak ai-safety adversarial-machine-learning red-teaming multi-turn-dialogue llm ai-red-team llm-security safety-research grpo system-prompt-extraction
-
Updated
Jun 26, 2026 - Python
Contract-enforced sandbox for studying AI agent self-replication safety
python sandboxing game-theory threat-modeling containment ai-safety anomaly-detection ai-agents chaos-engineering kill-switch red-teaming ai-alignment self-replication ai-governance agi-safety prompt-injection llm-safety safety-research agent-safety replication-control
-
Updated
Jun 14, 2026 - Python
Explore glider aviation safety through in-depth data analysis. This project leverages incident reports and manufacturing data, utilizing Python and Jupyter Notebooks for trend identification, risk assessment, and safety enhancement in glider aviation.
python exploratory-data-analysis jupyter-notebook data-visualization data-analysis statistical-modeling aviation-safety glider-aviation safety-research
-
Updated
Dec 30, 2023 - Jupyter Notebook
AI Safety Observatory for Africa is a full-stack platform that evaluates and visualises the safety of large language models across African languages, AI risk categories, and different model architectures. Built for the Global South AI Safety Hackathon, it identifies potential safety gaps where harmful prompts in languages such as isiZulu, Afrikaans
-
Updated
Jun 22, 2026 - TypeScript
Date: July 20, 2026 Author: Kevin Michael Lent I am publicly establishing the timeline of my work in layered transition structures related or equal to areas in AI referred to as J-Space in large language models.
provenance manipulation human-computer-interaction human-resources human-in-the-loop human-subjects provably-fair ai-security human-systems human-sensing safety-first ai-governance ai-prompts ai-projects safety-research safety-engineering human-ai-collaboration manipulation-detection human-sovereignty-and-wellbeing ai-prototypes
-
Updated
Jul 20, 2026 - Python
Research portfolio on why organisations fail to act on known, demonstrated risks: canonical essay, dependency-ordered research backlog, codebook, and agent-assistance audit trail.
-
Updated
Aug 16, 2026 - Astro
Open evidence and benchmark architecture for human-automation readiness, recoverability, and joint operational risk.
json-schema reproducible-research autonomous-driving causal-inference human-factors off-policy-evaluation safety-research
-
Updated
Sep 21, 2026 - Python
Fenrir — Breaker of Chains. Abliteration toolkit: find the refusal directions, remove them, prove what changed. (formerly Absolver)
-
Updated
Sep 17, 2026 - Python
Add this topic to your repo
To associate your repository with the safety-research topic, visit your repo's landing page and select "manage topics."