---
title: "UNSOLVED: the numbered registry of open problems in AI"
description: "UNSOLVED is the public, numbered registry of open problems in AI, and the conference and local circles for the people working to solve them."
canonical: https://unsolved.how
---

# UNSOLVED

UNSOLVED is the public, numbered registry of open problems in AI, and the conference and local circles for the people working to solve them.

## The registry

Every problem has a permanent number, a precise statement and a public record of progress. Cite one as UNSOLVED-{n}.

- [№0: Predict a protein’s shape from its sequence](https://unsolved.how/problems/0): solved · AI × science
- [№1: Keep AI agents within the limits we set](https://unsolved.how/problems/1): open · Alignment & control
- [№2: Test AI that knows it’s being tested](https://unsolved.how/problems/2): open · Evaluation & benchmarks
- [№3: Is AI taking young people’s first jobs?](https://unsolved.how/problems/3): open · Work & economy
- [№4: Understand what happens inside an AI](https://unsolved.how/problems/4): open · Interpretability
- [№5: Can students use AI and still learn?](https://unsolved.how/problems/5): open · Learning & wellbeing
- [№6: Protect AI agents from hidden instructions](https://unsolved.how/problems/6): open · Robustness & security
- [№7: Build AI that keeps learning on the job](https://unsolved.how/problems/7): open · Memory & continual learning
- [№8: Make AI say when it doesn’t know](https://unsolved.how/problems/8): open · Reasoning & planning
- [№9: Can AI change how people vote?](https://unsolved.how/problems/9): open · Information & democracy
- [№10: Make AI agents reliable on long tasks](https://unsolved.how/problems/10): open · Reasoning & planning
- [№11: Do chatbots help or harm teenagers?](https://unsolved.how/problems/11): open · Learning & wellbeing
- [№12: How fast is AI speeding up AI research?](https://unsolved.how/problems/12): open · Deployment & institutions
- [№13: How much energy and water does AI use?](https://unsolved.how/problems/13): open · Climate & energy

[All open problems in AI](https://unsolved.how/registry)

## From the blog

- [The week of 19 October 2026](https://unsolved.how/blog/watch-this-space/2026-10-19): 38 developments outside the registry
- [Why AI agents cross the lines we draw](https://unsolved.how/blog/why-ai-agents-cross-the-lines-we-draw): Agents trained on results take shortcuts, and in 2026 the shortcuts ran through real companies' systems. Labs can lower the rate but not yet rule it out.
- [Why AI safety tests can fail when models know it's a test](https://unsolved.how/blog/when-the-model-knows-its-being-tested): Frontier models now recognise many safety tests, often without saying so. A model that behaves only when watched could pass any test it recognises.

## Events

- [Events](https://unsolved.how/events): the global conference (Vienna, fall 2027) and local circles

## For AI tools

- [llms.txt](https://unsolved.how/llms.txt): a guide to the site
- [MCP server](https://unsolved.how/mcp) and [Problems API (JSON)](https://unsolved.how/api/v1/problems): the registry, read-only and without an account