Skip to content

@safety-research Safety Research

Change the repository type filter

All

    Repositories list

    45 repositories

    • legibility

      Public
      Which models are illegible under what conditions, and why? How does that impact monitorability?
      Jupyter Notebook
      •
      MIT License
      •0•0•0•0•Updated Apr 19, 2026Apr 19, 2026

    • aligning-ai-orgs

      Public
      Python
      •0•0•0•0•Updated Apr 17, 2026Apr 17, 2026

    • introspection-mechanisms

      Public
      introspection mechanisms
      Python
      •5•17•0•0•Updated Apr 16, 2026Apr 16, 2026

    • petri

      Public
      An alignment auditing agent capable of quickly exploring alignment hypothesis
      Python
      •
      MIT License
      •149•995•3•4•Updated Apr 16, 2026Apr 16, 2026

    • auditing-agents

      Public
      Python
      •2•13•1•2•Updated Apr 14, 2026Apr 14, 2026

    • automated-w2s-research

      Public
      Python
      •29•136•0•0•Updated Apr 13, 2026Apr 13, 2026

    • crosscoder_emergent_misalignment

      Public
      Applying crosscoder model diffing to emergently misaligned models
      Python
      •0•5•55•8•Updated Apr 1, 2026Apr 1, 2026

    • agent-transcript-editor

      Public
      Web UI for viewing, editing, and AI-assisted red teaming of AI agent transcripts
      Python
      •0•1•0•0•Updated Mar 31, 2026Mar 31, 2026

    • trusted-monitor

      Public
      Evaluate AI agent transcripts for suspicious behavior (0-100 scoring)
      Python
      •0•1•0•0•Updated Mar 28, 2026Mar 28, 2026

    • safety-tooling

      Public
      Inference API for many LLMs and other useful tools for empirical research
      Python
      •
      MIT License
      •36•115•13•18•Updated Mar 23, 2026Mar 23, 2026

    • PurpleLlama

      Public
      Set of tools to assess and improve LLM security.
      Python
      •
      Other
      •723•0•0•0•Updated Feb 23, 2026Feb 23, 2026

    • bloom

      Public
      bloom - evaluate any behavior immediately  🌸🌱
      Python
      •
      MIT License
      •162•1.3k•0•8•Updated Feb 17, 2026Feb 17, 2026

    • casr

      Public
      Collect crash (or UndefinedBehaviorSanitizer error) reports, triage, and estimate severity.
      Rust
      •
      Apache License 2.0
      •36•0•0•0•Updated Feb 3, 2026Feb 3, 2026

    • assistant-axis

      Public
      The Assistant Axis is a direction in activation space that captures how "Assistant-like" a model's behavior is. Models can drift away from the Assistant during …
      Jupyter Notebook
      •36•127•2•1•Updated Jan 20, 2026Jan 20, 2026

    • selective-gradient-masking

      Public
      Training Transformers with knowledge localization (SGTM)
      Python
      •
      MIT License
      •5•51•0•0•Updated Jan 11, 2026Jan 11, 2026

    • how-ai-impacts-skill-formation

      Public
      Repo for measuring whether using AI tools inhibits skill formation and development
      Python
      •2•13•0•1•Updated Jan 3, 2026Jan 3, 2026

    • A3

      Public
      Python
      •
      Apache License 2.0
      •1•14•0•0•Updated Dec 29, 2025Dec 29, 2025

    • inverse-scaling-ttc

      Public
      Inverse Scaling in Test-Time Compute
      Python
      •
      MIT License
      •2•25•0•0•Updated Dec 3, 2025Dec 3, 2025

    • impossiblebench

      Public
      Official Inspect Implementation for "ImpossibleBench: Measuring LLMs' Propensity of Exploiting Test Cases"
      Python
      •
      MIT License
      •8•37•0•0•Updated Dec 1, 2025Dec 1, 2025

    • SCONE-bench

      Public
      MIT License
      •29•177•5•0•Updated Nov 25, 2025Nov 25, 2025

    • unsupervised-truth-probes

      Public
      Python
      •0•5•1•0•Updated Nov 24, 2025Nov 24, 2025

    • ciphered-reasoning-llms

      Public
      Jupyter Notebook
      •
      MIT License
      •15•9•0•0•Updated Nov 20, 2025Nov 20, 2025

    • safety-examples

      Public
      Jinja
      •
      MIT License
      •15•25•0•2•Updated Nov 11, 2025Nov 11, 2025

    • weight-steering

      Public
      Python
      •3•8•0•0•Updated Nov 11, 2025Nov 11, 2025

    • misalignment-scraper

      Public
      AI misalignment detection and reproduction tool for social media content
      Python
      •
      MIT License
      •0•1•0•0•Updated Nov 1, 2025Nov 1, 2025

    • believe-it-or-not

      Public
      Code and data for editing model beliefs with SDF and other methods, and for evaluating the depth of the implanted beliefs.
      Python
      •
      MIT License
      •4•13•1•0•Updated Oct 23, 2025Oct 23, 2025

    • science-synth-facts

      Public
      Python
      •5•6•1•0•Updated Oct 22, 2025Oct 22, 2025

    • finetuning-auditor

      Public
      Auditing agents for fine-tuning safety
      Python
      •
      MIT License
      •3•20•0•0•Updated Oct 21, 2025Oct 21, 2025

    • inoculation-prompting

      Public
      Python
      •5•10•0•0•Updated Oct 13, 2025Oct 13, 2025

    • verl

      Public
      veRL: Volcano Engine Reinforcement Learning for LLM
      Python
      •
      Apache License 2.0
      •3.7k•4•0•1•Updated Oct 3, 2025Oct 3, 2025
    ProTip! When viewing an organization's repositories, you can use the props. filter to filter by custom property.