Search Anthropic

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.
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.
Claude’s values across models and languages
New research into how the values expressed by Claude vary across the top 20 languages on Claude.ai.
An off switch for dual-use knowledge in AI models
New results on a method of controlling access to potentially dangerous AI capabilities
New research on how we've reduced agentic misalignment.
How people ask Claude for personal guidance
In a new study, we looked at what types of guidance people ask of Claude. We explored how Claude responded across different domains, focusing particularly on how rates of excessive validation or praise varied by the topic of guidance.
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.
Claude's new constitution
A new approach to a foundational document that expresses and shapes who Claude is
Claude's Constitution
A detailed description of our vision for Claude’s values and behavior.
From shortcuts to sabotage: Natural emergent misalignment from reward hacking
In the latest research from Anthropic’s alignment team, we show for the first time that realistic AI training processes can accidentally produce misaligned models.
Measuring political bias in Claude
We introduce and open-source an automated evaluation for political even-handedness, testing six models across 1,350 prompt pairs.
Expanding our use of Google Cloud TPUs and Services
Announcing a dramatic increase in Anthropic's compute resources through an expanded partnership with Google Cloud.
A small number of samples can poison LLMs of any size
In a joint study with the UK AI Security Institute and the Alan Turing Institute, we found that as few as 250 malicious documents can produce a "backdoor" vulnerability in a large language model—regardless of model size or training data volume.
Rahul Patil joins Anthropic as Chief Technology Officer
Rahul Patil brings 20+ years of infrastructure experience from Stripe, AWS, Microsoft, and Oracle Cloud to lead engineering at Anthropic.
Updates to Consumer Terms and Privacy Policy
Claude Free, Pro, and Max users can now opt in to having conversations used to improve models, with a five-year retention period for participants.
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.
Reasoning models don't always say what they think
Research from Anthropic on the faithfulness of AI models' chain-of-thought reasoning.
Alignment faking in large language models
A paper from Anthropic's Alignment Science team on Alignment Faking in AI large language models
Developing a computer use model
Developing a computer use model
Fine-tune Claude 3 Haiku in Amazon Bedrock
Claude 3 Haiku can now be fine-tuned in Amazon Bedrock with custom training data for faster, more accurate performance at lower cost.
Sycophancy to subterfuge: Investigating reward tampering in language models