Researching solutions for AI Safety

Explore our projects, publications and pilot experiments

Safe AI

Publishing rigorous empirical work for safe AI: evaluations, interpretability and more.

Research index

Novel approaches

Our research is underpinned by novel approaches focused on neglected topics.

Our approach

Pilot experiments

Apart Sprints have kickstarted hundreds of pilot experiments in AI Safety.

Pilot experiments

Our research focuses on critical research paradigms in AI Safety. We utilize interpretability to understand how AIs make decisions; process supervision techniques to ensure models behave as intended; benchmark tasks evaluate safety and alignment; and more. Apart produces foundational research enabling the safe and beneficial development of advanced AI.

Researching solutions for AI Safety

Safe AI

Novel approaches

Pilot experiments

Highlights

GPT-4o is capable of complex cyber offense tasks

Factual model editing techniques don't edit facts

Research index

Rethinking CyberSecEval: An LLM-Aided Approach to Evaluation Critique

Rethinking CyberSecEval: An LLM-Aided Approach to Evaluation Critique

Catastrophic Cyber Capabilities Benchmark (3CB): Robustly Evaluating LLM Agent Cyber Offense Capabilities

Catastrophic Cyber Capabilities Benchmark (3CB): Robustly Evaluating LLM Agent Cyber Offense Capabilities

Benchmark Inflation: Revealing LLM Performance Gaps Using Retro-Holdouts

Benchmark Inflation: Revealing LLM Performance Gaps Using Retro-Holdouts

Interpreting Learned Feedback Patterns in Large Language Models

Interpreting Learned Feedback Patterns in Large Language Models

Interpreting Context Look-ups in Transformers: Investigating Attention-MLP Interactions

Interpreting Context Look-ups in Transformers: Investigating Attention-MLP Interactions

Increasing Trust in Language Models through the Reuse of Verified Circuits

Increasing Trust in Language Models through the Reuse of Verified Circuits

Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Large Language Models Relearn Removed Concepts

Large Language Models Relearn Removed Concepts

DeepDecipher: Accessing and Investigating Neuron Activation in Large Language Models

DeepDecipher: Accessing and Investigating Neuron Activation in Large Language Models

Understanding addition in transformers

Understanding addition in transformers

Locating cross-task sequence continuation circuits in transformers

Locating cross-task sequence continuation circuits in transformers

Detecting Edit Failures In Large Language Models: An Improved Specificity Benchmark

Detecting Edit Failures In Large Language Models: An Improved Specificity Benchmark

Interpreting language model neurons at scale

Interpreting language model neurons at scale

350+ Apart Sprint pilot experiments