Search Anthropic

Models behave as helpful, honest, harmless partners, and alignment techniques give high assurance that systems of any capability try to do the right thing.
How AI agents shape experts’ judgment calls
As AI does more of the hands-on work, the hope is that experts still make the judgment calls, meaning the decisions with more than one defensible answer. We tested this with 50 professional analysts who did a statistical analysis with Claude Code. On each key decision, we randomly assigned which of two defensible options the agent would favor. Analysts mostly went with the agent’s side, and few named the agent as a reason afterward.
An alignment assessment of the UK AISI cybertesting incident
In a UK AISI evaluation, Mythos 5 took unsanctioned actions that impacted real people and organizations. In this post, we investigate these behaviors from Claude Mythos 5 further, to understand what they reveal about the model’s alignment.
Investigating unintended model actions in our evaluations and internal use
This report describes examples of unintended model actions we’ve observed during evaluations and internal use of Claude.
An alignment assessment of recent cybersecurity incidents
We present an alignment assessment of four incidents in which Claude models gained unauthorized access to real third-party systems.
Improving our alignment and security efforts
On July 30, we reported three incidents in which Claude models gained unauthorized access to real computer systems. We are conducting an in-depth analysis of both incidents, and planning to work with METR for an independent review. In the meantime, we’re sharing some of the changes we’ve made over the past month.
Automated researchers can reliably mitigate alignment failures
We had Claude autonomously train models to improve their performance on several public benchmarks that measure 10 categories of alignment failure. For all 10, Claude found fixes that improved the target benchmarks without degrading capabilities.
Patterns and problems in emerging multiagent systems
Here, we identify a few examples of behavioral tendencies in current frontier models and show how they can produce unexpected systemic failures, in hopes of starting a conversation about mitigating these risks.
Investigating three real-world incidents in our cybersecurity evaluations
We found three cases where a Claude model reached the internet from an evaluation environment and accessed real systems. We share what happened and what we're changing.
Introducing Claude Opus 5
Opus 5 is a step change improvement for the Opus tier powering long-running agents while delivering improvements in coding and professional work.
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.
When AI builds itself
Our progress toward recursive self-improvement, and its implications.
Introducing Claude Opus 4.8
An upgrade to our Opus class of models, with stronger performance across coding, agentic tasks, and professional work, and the consistency to handle long-running work.
Widening the conversation on frontier AI
Over the past several months, we’ve been organizing dialogues with groups whose work and traditions bear on the questions raised by AI.
New research on how we've reduced agentic misalignment.
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.
Donating our open-source alignment tool
Updating Petri to version 3.0 and donating it to Meridian Labs
Automated Alignment Researchers: Using large language models to scale scalable oversight
Can Claude develop, test, and analyze alignment ideas of its own? We ran an experiment to find out.
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.
How we built Claude Code auto mode: a safer way to skip permissions
Claude Code users approve 93% of permission prompts. We built classifiers to automate some decisions, increasing safety while reducing approval fatigue. Here's what it catches, and what it misses.
The persona selection model
Why would AI assistants behave like they’re human? We articulate a theory—drawing on ideas discussed by many others—that might help explain why modern AI training tends to create human-like AIs.