Alignment faking in large language models

uk ai safety summit

specific versus general principles for constitutional ai

towards understanding sycophancy in language models

collective constitutional ai aligning a language model with public input

decomposing language models into understandable components

towards monosemanticity decomposing language models with dictionary learning

evaluating ai systems

anthropic amazon

prompting long context

anthropics responsible scaling policy

the long term benefit trust

anthropic bcg

claude pro

skt partnership announcement

releasing claude instant 1 2

influence functions

studying large language model generalization with influence functions

frontier threats red teaming for ai safety

frontier model security

measuring faithfulness in chain of thought reasoning