Database
The AI Safety database is a work in progress, built from public sources. Missing, or out of date? You can list your project yourself. Feel free to send us ideas or feedback!
| Name | Type | Tags | Endorsed | Team | Raising $ |
|---|
| Name | Type | Tags | Endorsed | Team | Raising $ |
|---|
Showing 101-117 of 117·Sorted by endorsements
| Name | Type | Tags | Endorsed | Team | Raising $ |
|---|
| Project | Research | - | - | ||
| Project |
Individual |
| - |
| - |
| Project | Individual | - | - |
| Project | Individual | - | - |
| Project | Individual | - | - |
| Project | Research | - | - |
| Project | Research | - | - |
| Project | Research | - | - |
| Project | Research | - | - |
| Project | Research | - | - |
| Project | Research | - | - |
| Project | Research | - | - |
| Project | Research | - | - |
| Project | Research | - | - |
| Org | - | - | 3 |
| Project | Community | - | - |
| Org | Funder | - | 15 |
Codes 121 top comments on a widely read essay about a July 2026 AI incident, finding commenters who adopted its anthropomorphic framing rarely named any possible action.
Testing whether AI agents can comply with corrective instructions after their earlier actions have already reduced or eliminated effective human control.
Travel support for an independent researcher from Bangladesh to present his single-author NeurIPS 2026 paper on how efficiently LLMs reason.
A model-agnostic method and tool for independently testing whether advanced AI systems remain stable and appropriately corrigible without requiring access to model internals.
Independently replicate and open-source a model-free AI agent monitor that caught 209 of 210 sabotage runs, plus static checks that find fail-open defects in safety tooling.
Introduces Governance Drift and a reproducible framework for testing whether AI models maintain governance-aware behavior under user pressure across jurisdictions and domains.
Develops a behavioral comparison protocol using independently generated solutions and multi-round interrogation to detect injected backdoors that evade weak-model code monitoring.
Tests whether Qwen2.5-7B-Instruct internally represents Colombian identity, socioeconomic status, and stereotypes from Spanish and English prompts using layer-20 activation verbalization.
Tests whether sentence embeddings can recover a poisoned dataset’s hidden principal without clean data or a model, and maps the limits of this defense.
A non-LLM graph neural network compares code property graphs with trusted references to detect backdoors at low false-positive rates and provide node-level audit trails.
SpecCheck detects intent drift in LLM-formalized requirements using ambiguity elicitation and cross-model validation, evaluated on Lean 4 theorems and Python property-based tests.
The project evaluates secret-loyalty detections with matched controls and finds prior signals spurious, while verifying that organism C is byte-identical to the base model.
Develops an internal-representation scanning tool to detect hidden loyalties in AI models by comparing entity representations, and critiques existing sleeper-agent benchmarks.
An open-source multilingual safety benchmark evaluates Llama models across South and Southeast Asian languages, revealing under- and over-refusal caused by English-centric safety training.
Agentic AI alignment research collaboration — inspired by fundamental principles from biology and economics
Building Hispanic participation in advanced AI safety and helping Hispanic-serving institutions advocate for effective AI governance.
An independent Swiss foundation launched in February 2024, spun out of NTI | bio, that develops free open-source tools for DNA synthesis screening and works to strengthen international biosecurity norms. Led by Piers Millett, IBBIS created the Common Mechanism (commec) and runs consortia to support international standards for synthesis screening, including the SBRC.
