Search

44 results

Interpretability
Jul 6, 2026
A global workspace in language models
New interpretability research reveals an emergent mental workspace in Claude that holds internal thoughts that don’t appear in the model’s output.
View article
Announcements
May 25, 2026
Anthropic co-founder Chris Olah's remarks on Pope Leo XIV's encyclical "Magnifica humanitas"
The full text of Chris Olah's remarks on the Pope's encyclical on AI.
View article
Interpretability
May 7, 2026
Natural Language Autoencoders: Turning Claude’s thoughts into text
AI models like Claude talk in words but think in numbers. In this study, we train Claude to translate its thoughts into human-readable text.
View article
Interpretability
Apr 2, 2026
Emotion concepts and their function in a large language model
All modern language models sometimes act like they have emotions. What’s behind these behaviors? Our interpretability team investigates.
View article
Interpretability
Mar 13, 2026
A “diff” tool for AI: Finding behavioral differences in new models
"Model diffing" is a powerful way to understand how models change during fine-tuning. Our new research paper extends this method to its most challenging and general use case: comparing models with entirely different architectures.
View article
Interpretability
Jan 19, 2026
The assistant axis: situating and stabilizing the character of large language models
In a new paper, we look at several open-weights language models, map out how their neural activity defines a “persona space,” and situate the “Assistant” persona within that space.
View article
Alignment
Jan 9, 2026
Next-generation Constitutional Classifiers: More efficient protection against universal jailbreaks
Constitutional Classifiers++ improve on our previous approach, yielding a system that is even more robust, has a much lower refusal rate, and is dramatically cheaper to run.
View article
Interpretability
Oct 29, 2025
Signs of introspection in large language models
Our new research provides evidence for some degree of introspective awareness in our current Claude models, as well as a degree of control over their own internal states.
View article
Interpretability
Aug 1, 2025
Persona vectors: Monitoring and controlling character traits in language models
In a new paper, we identify patterns of activity within an AI model’s neural network that control its character traits, which we call persona vectors.
View article
Interpretability
May 29, 2025
Open-sourcing circuit tracing tools
We recently introduced a new method to trace the thoughts of a large language model. Today, we’re open-sourcing the method so that anyone can build on our research.
View article
Interpretability
Mar 27, 2025
Tracing the thoughts of a large language model
Today, we're sharing two new papers that represent progress on the development of an “AI microscope,” and the application of it to see new “AI biology.”
View article
Alignment
Mar 13, 2025
Auditing language models for hidden objectives
A collaboration between Anthropic's Alignment Science and Interpretability teams
View article
Interpretability
Feb 20, 2025
Insights on crosscoder model diffing
Preliminary interpretability work on crosscoder model diffing, a method for understanding how two models differ.
View article
Societal Impacts
Oct 25, 2024
Evaluating feature steering: A case study in mitigating social biases
Our preliminary (mixed) findings on how feature steering might play a role in creating safer model outputs.
View article
Interpretability
Oct 16, 2024
Using dictionary learning features as classifiers
An interpretability update on using dictionary learning features as classifiers for detecting content or behaviors in model activations.
View article
Interpretability
Oct 1, 2024
Circuits Updates – September 2024
Preliminary interpretability notes on successor heads and on oversampling topics in sparse autoencoder training for safety-relevant features.
View article
Interpretability
Sep 6, 2024
Circuits Updates – August 2024
Preliminary interpretability notes on quantified autointerpretability evaluations for sparse autoencoders and self-explaining features.
View article
Interpretability
Jul 31, 2024
Circuits Updates – July 2024
Five open challenges for mechanistic interpretability, plus notes on linear representations, rare features, and measuring feature sensitivity.
View article
Interpretability
Jun 28, 2024
Circuits Updates – June 2024
Comparing alternative sparse autoencoder architectures on the sparsity and reconstruction tradeoff, plus notes on external interpretability work.
View article
Interpretability
Jun 13, 2024
The engineering challenges of scaling interpretability
Our researchers reflect on the close relationship between scientific and engineering progress, and discuss the technical challenges they encountered in scaling our interpretability research to much larger AI models.
View article
Interpretability
May 23, 2024
Golden Gate Claude
When we amplify the "Golden Gate Bridge" feature, Claude's responses focus on the bridge. For a short time, anyone can interact with this model.
View article
Interpretability
May 21, 2024
Mapping the mind of a large language model
We have identified how millions of concepts are represented inside Claude Sonnet. This is the first ever detailed look inside a modern, production-grade large language model.
View article
Interpretability
Apr 26, 2024
Circuits Updates – April 2024
Preliminary interpretability notes on scaling laws for dictionary learning, sparse autoencoder training, and circuit-analysis techniques.
View article
Alignment
Apr 23, 2024
Simple probes can catch sleeper agents
Linear classifiers trained on model activations detect when sleeper-agent models are about to defect, with AUROC above 99%.
View article
Interpretability
Mar 8, 2024
Reflections on qualitative research
Why qualitative research plays a central role in interpretability, with heuristics for judging the quality of qualitative work.
View article
Interpretability
Oct 5, 2023
Decomposing language models into understandable components
We outline evidence that there are better units of analysis than individual neurons, and we have built machinery that lets us find these units in small transformer models.
View article
Interpretability
Oct 5, 2023
Towards monosemanticity: Decomposing language models with dictionary learning
In our latest paper, we outline evidence that there are better units of analysis than individual neurons, and we have built machinery that lets us find these units in small transformer models.
View article
Alignment
Aug 8, 2023
Studying large language model generalization with influence functions
This paper scales influence functions to 52B parameter models, revealing how generalization patterns become more abstract with model size.
View article
Alignment
Aug 8, 2023
Tracing model outputs to the training data
Scaling influence functions to 52B-parameter models shows generalization becomes more abstract and cross-lingual as models grow.
View article
Alignment
Jul 18, 2023
Measuring faithfulness in Chain-of-Thought reasoning
Does chain-of-thought reasoning reflect a model's actual decision process? We find faithfulness varies by task and declines in larger models.
View article
Policy
Jun 13, 2023
Charting a path to AI accountability
Anthropic's NTIA submission recommends rigorous model evaluations, pre-registration of large training runs, and mandatory external red teaming.
View article
Interpretability
May 24, 2023
Interpretability dreams
An informal essay on Anthropic's long-term vision for mechanistic interpretability, starting with solving superposition.
View article
Interpretability
May 24, 2023
Circuits Updates — May 2023
Developing interpretability ideas on dictionary learning, defining features, and the possibility of superposition in attention heads.
View article
Interpretability
May 4, 2023
Distributed representations: Composition & superposition
Since publishing our original paper on superposition, we've had more time to reflect on the relationship between the topics and discuss it with people, and wanted to expand on our earlier discussion in the related work section and share a few thoughts.
View article
Interpretability
Mar 16, 2023
Privileged bases in the transformer residual stream
Our mathematical theories of the Transformer architecture suggest that individual coordinates in the residual stream should have no special significance. Recent work has shown that this observation is false in practice. We investigate this phenomenon and provisionally conclude that the per-dimension normalizers in the Adam optimizer are to blame for the effect.
View article
Policy
Mar 8, 2023
Core views on AI safety: When, why, what, and how
Why we believe AI safety research is critical, and how we approach it at Anthropic.
View article
Interpretability
Jan 5, 2023
Superposition, memorization, and double descent
A toy model proves to be a surprisingly rich case study for overfitting.
View article
Interpretability
Sep 14, 2022
Toy models of superposition
In this paper, we use toy models to investigate how and when models represent more features than they have dimensions.
View article
Interpretability
Jun 17, 2022
Softmax Linear Units
We report an architectural change which appears to substantially increase the fraction of MLP neurons which appear to be "interpretable," at little to no cost to ML performance.
View article
Interpretability
May 21, 2022
Scaling laws and interpretability of learning from repeated data
In this paper, we attempt to study repeated data systematically and to understand its effects mechanistically.
View article
Announcements
Apr 29, 2022
Anthropic Raises Series B to build steerable, interpretable, robust AI systems
Anthropic raised $580 million in Series B funding to build infrastructure for studying and improving the safety of large-scale AI models.
View article
Interpretability
Mar 8, 2022
In-context learning and induction heads
In this paper, we present preliminary and indirect evidence for a tantalizing hypothesis: that induction heads might constitute the mechanism for the actual majority of all in-context learning in large transformer models.
View article
Interpretability
Dec 22, 2021
A Mathematical Framework for Transformer Circuits
A framework for reverse engineering transformers, revealing how attention heads work as independent operations.
View article
Announcements
May 28, 2021
Anthropic raises $124 million to build more reliable, general AI systems
Anthropic's Series A, led by Jaan Tallinn, funds research toward steerable, interpretable, and robust large-scale AI systems.
View article