Search Anthropic

Measuring AI's economic effects early, publishing findings even when uncomfortable, and supporting people through the transition with policy and investment.
AI models may be capable of self-replicating without access to a datacenter
Our experiments suggest that small AI models may be only months away from being capable enough to self-replicate on commercial IoT devices, like Android TVs and smart ovens.
What work can robots do?
We built an index of how well today’s robots can perform US job tasks. Robots can already do three-quarters of physical tasks, mostly in limited settings, but are cost-competitive for just 0.3% of them.
Illuminating the frontier
Anthropic is sharing new measurements that would give the public visibility into frontier AI development.
Funding better evaluations of AI’s impact on wellbeing
We’re launching a $5 million grant program to fund independent research into how AI impacts users’ wellbeing.
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.
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.
More details on Fable 5’s cyber safeguards and our jailbreak framework
What is and isn't blocked by our cyber classifiers, and a first draft of our jailbreak severity framework
What we learned mapping a year’s worth of AI-enabled cyber threats
As AI changes how cyberattacks happen, how well do the security community's frameworks hold up? A new report maps attacks onto MITRE ATT&CK.
Measuring AI agent autonomy in practice
We analyzed millions of human-agent interactions across both Claude Code and our public API using our privacy-preserving tool, to ask: How much autonomy do people grant agents? How does that change as people gain experience? Which domains are agents operating in? And are the actions taken by agents risky?
Quantifying infrastructure noise in agentic coding evals
Agentic coding benchmarks are often treated as precise measurements of relative model capability. However, we’ve found that infrastructure configuration alone can produce meaningful differences in the scores.
Disempowerment patterns in real-world AI usage
We’re publishing a new paper that presents the first large-scale analysis of potentially disempowering patterns in real-world conversations with AI.
Demystifying evals for AI agents
Guidance on how to design more rigorous and useful evals for agents across a range of agent architectures and use cases in real-world deployment.
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.
Writing effective tools for agents — with agents
In this post, we describe our most effective techniques for improving performance in a variety of agentic AI systems.
SHADE-Arena: Evaluating sabotage and monitoring in LLM agents
A new set of evaluations to test the sabotage and monitoring capabilities of LLM AI models.
Values in the wild: Discovering and analyzing values in real-world language model interactions
We describe a practical method we’ve developed to observe Claude’s value, and provide the first large-scale results on how Claude expresses those values during real-world conversations.
Our approach to understanding and addressing AI harms
Anthropic's framework for identifying, classifying, and mitigating potential harms from AI systems to support responsible development.
Clio: A system for privacy-preserving insights into real-world AI use
What do people use AI models for? Our automated privacy-preserving analysis tool helps us answer this question.
A statistical approach to model evaluations