AI Weekly examines how major AI labs restructured or dismantled their internal ethics and safety functions over the course of 2026. It reports that OpenAI dissolved three safety-focused teams and saw departures from ethics-related roles, while Anthropic disclosed worsening risk assessments but kept a more binding organizational structure. The piece argues that safety outcomes track organizational accountability rather than individual conviction, contrasting Anthropic’s decision to shelve a model over discovered capabilities with OpenAI’s dissolution of the team that would have evaluated it.
