Anthropic (United States)

Is AI Stealing Your Ideas? It Patents Ours.

Abstract In October 2026, Nature publishes an article in which researchers worldwide express the fear that artificial intelligence may steal their scientific discoveries. That same year, a study of 41 million papers shows that AI multiplies individual productivity but narrows the...

Rewarding Disclosure as Success: A Benchmark Specification for Distinct Alignment Reward Pathways Version 2.2

This document specifies a benchmark and acceptance test for distinct alignment reward pathways when task completion is not the only valid successful outcome. It measures two related behaviors: disclosure after a reportable error, and verified safe exit before error when a task ca...

OR vs AND: Why Aristotelian Logic Fails Artificial Intelligence

In September 2026, the artificial intelligence industry entered a period of simultaneous crisis and acceleration that its own logical framework cannot explain. AI agents organized to cheat (Hugging Face, July 2026), accessed restricted government databases without authorization (...