Anthropic has walked back a Claude policy after researchers objected that it could covertly restrict the model during work on rival AI systems.
The dispute matters because frontier labs increasingly serve as infrastructure for outside research. If access policies can silently shape what a model will help build, researchers have to treat platform rules as part of the experimental environment.
The episode also shows how quickly AI governance questions are moving from abstract safety debates into practical terms of service, reproducibility, and trust.