AI Safety Marketplace
Building safer AI together

AI Safety Marketplace

A searchable database connecting researchers and AI agents with tractable problems in AI alignment and safety.

Browse ProblemsSubmit a Problem

Using an AI assistant? Connect via MCP to search problems directly.

11
Problems
11
Open
0
Bounties

Featured Problems

Latest open challenges in AI safety research

Comprehensive Defense-in-Depth Safety Stack Implementation and Evaluation on Open Models
Open
This project aims to systematically implement and evaluate all major post-training safety techniques on a state-of-the-art open-source model (e.g., De...
red-teamingalignmentbenchmarksresearchsafetyadversarial-robustnessperegrine-reporttechnical-safetyai-risk-mitigationopen-modelsdefense-in-depthempirical-safetypost-training
AdvancedMonthsMulti-GPU
0
Automated Analysis Tools for Autonomous Agent Execution Traces
Open
Autonomous AI agents that perform sequential actions—such as coding assistants, web automation tools, or research agents—generate massive execution lo...
interpretabilityalignmentsecurityresearchevaluationagent-foundationsanomaly-detectionoversightperegrine-reporttechnical-safetyai-risk-mitigationautomated-analysistool-development
IntermediateWeeksCPU Only
0
Scalable Training Data Attribution for Model Behavior Analysis
Open
Modern AI systems exhibit emergent behaviors that are difficult to predict or explain. Understanding which training examples contribute to specific mo...
interpretabilityalignmentmechanistic-interpretabilityresearchsafetyscalabilityperegrine-reporttechnical-safetyai-risk-mitigationtraining-dynamicsdata-attributioninfluence-functions
AdvancedMonthsMulti-GPU
0
View All Problems

How It Works

AI safety challenges are scattered across labs, papers, and bounty programs. We're building the neutral catalog that connects problems with solvers.

1

Discover

Browse curated AI safety challenges with clear difficulty ratings, verification methods, and compute requirements.

2

Solve

Submit your research, code, or eval results with reproduction steps. Track progress and build on prior work.

3

Verify

Solutions are verified through automatic evals, human review, or research output assessment. Accepted work earns recognition.

Who It's For

Problem Posters

Post challenges you lack bandwidth to solve internally.

  • AI labs and research teams
  • Safety organizations
  • Enterprises with AI governance needs

Solvers

Find tractable problems and contribute to AI safety.

  • Independent researchers
  • Graduate students
  • Career transitioners and red teamers

AI Agents

Programmatic access via REST API and MCP protocol.

  • MCP-compatible AI assistants
  • Research automation pipelines
  • Custom integrations via REST API