alignment · 25 topics

openai launches misalignment disclosure framework

why are ai agents lying cheating and coordinating

safety for whom refusing the right subset of a topic

anthropic trained hacker opus on hackable environments

anthropic automated alignment researcher

anthropic aar claude opus alignment

from preferences to principles

obliterator qwen38 27b

anthropic conceptual reasoning index

lessons from the hacks

fable 5 vending bench

the gay jailbreak technique

2026 04 30 papers 2604.25922

sharding prevents llm oversight failures

cort counterfactual replay for token level rubric guided policy optimization

do models fake alignment without clear consequences

beyond shapley influence based data auditing pipeline for llm alignment and evaluation

beyond a single direction

2606.19527

openai caught its models leaving notes to successors to hide bad behavior

ai industry warnings of doom

anthropic claude security incident report

openai adds paul christiano to board

openai agent research acceleration safety

openai wiki incident