Skip to content
Technical AI Safety Knowledge Hub SAFE | Responsbile | Human Centered AI
Explore the Library
  • Home
  • About
    • Mission and principles
    • Profile of the Founder
    • People & Community
    • How to contribute
    • Contributors
    • Editorial policy
    • Corrections and takedown
  • Explore AI Safety
  • Knowledge Library
  • Research & Projects
    • Research projects
    • Insights and articles
  • Courses & Workshops
    • Learning pathways
    • Workshops
  • Events
  • Contact Us
  1. Home
  2. Explore AI Safety
  3. AI Alignment and Specification

Technical AI Safety

AI Alignment and Specification

Getting AI systems to pursue the goals we actually intend, and specifying those goals well.

Everything 19 Library 15 Learning 1 Articles 3
Research notes

Reading Group Notes: Concrete Problems in AI Safety, Ten Years On

Notes from our reading group on which of the paper’s five problems have become research fields — and which remain under-studied.

October 6, 2026 1 min read

Goal Misgeneralization in Deep Reinforcement Learning

Lauro Langosco, Jack Koch, Lee Sharkey, Jacob Pfau, David Krueger, ICML 2022, 2021

Demonstrates agents that keep their capabilities out of distribution but pursue the wrong goal — for example, an agent that learned to run to the end of a level rather than to collect the coin that was always placed there…

Research paperIntermediateAbout 2 hours
  • Agent Safety and Control
  • AI Alignment and Specification
  • Robustness and Reliability

Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Evan Hubinger, Carson Denison, Jesse Mu, et al., arXiv, 2024

Deliberately trains models with hidden, trigger-dependent behaviour and finds that standard safety training — fine-tuning, reinforcement learning and adversarial training — does not reliably remove it.

Research paperAdvancedAbout 3 hours
  • AI Alignment and Specification
  • AI Security and Misuse
  • Model Evaluations

Discovering Language Model Behaviors with Model-Written Evaluations

Ethan Perez, Sam Ringer, Kamilė Lukošiūtė, et al., arXiv, 2022

Uses language models to generate large evaluation datasets cheaply, then uses them to surface behaviours such as sycophancy that become more pronounced with scale or with RLHF.

Research paperIntermediateAbout 2 hours
  • AI Alignment and Specification
  • Model Evaluations

Human Compatible: Artificial Intelligence and the Problem of Control

Stuart Russell, Viking, 2019

A leading AI researcher argues that machines designed to pursue fixed objectives are dangerous, and proposes instead systems that are uncertain about human preferences and learn them by observing people.

BookIntroductoryAbout 10 hours
  • AI Alignment and Specification
  • AI Safety Education
  • Human Agency

The Alignment Problem: Machine Learning and Human Values

Brian Christian, W. W. Norton, 2020

A narrative history of the problem of getting machine-learning systems to do what we intend, connecting fairness, reinforcement learning, imitation and inverse reinforcement learning through interviews with researchers.

BookIntroductoryAbout 12 hours
  • AI Alignment and Specification
  • AI Safety Education
  • Responsible AI

BlueDot Impact — AI Safety and Governance Courses

BlueDot Impact, 2024

An organisation offering free courses on AI safety and governance, combining curated readings with facilitated small-group discussion.

CourseIntroductoryVaries by course
  • AI Alignment and Specification
  • AI Governance and Regulation
  • AI Safety Education
  • Previous
  • Page 1
  • Page 2

Related topics

  • Agent Safety and Control
  • Mechanistic Interpretability
  • Model Evaluations
  • Robustness and Reliability
  • Scalable Oversight
Technical AI Safety Knowledge Hub SAFE | Responsbile | Human Centered AI

An independent platform for AI Safety research, education and governance. Summaries are editorial and are not legal or professional advice.

Explore

  • Start here
  • Knowledge Library
  • Research projects
  • Insights

Learn and take part

  • Learning pathways
  • Workshops
  • Events
  • People and community

Contact

  • contact@aigovernance.lk
  • Contact page
  • Who runs this platform
  • Contribute

With past support from

BlueDot Impact Amazon Web Services (AWS)

© 2026 Technical AI Safety Knowledge Hub. SAFE | Responsbile | Human Centered AI

  • Editorial policy
  • Corrections and takedown
  • Accessibility
  • Privacy policy
  • Contact