OpenAI and Anthropic Reportedly Sought Control Over Safety Review Publications
A Reuters report indicates that OpenAI and Anthropic explored imposing stricter controls on third-party safety reviewers to prevent the disclosure of trade secrets. The companies are concerned about potential leaks from independent evaluations, which could compromise their intellectual property. OpenAI stated it typically redacts confidential information in its reports and allows evaluators to disclose if significant redactions were made. Meanwhile, Anthropic reportedly clarified that partners can manage their own review and publishing processes independently. This development highlights a growing tension within the AI industry over transparency versus proprietary protection. The situation underscores broader debates about how much control leading AI firms should have over independent assessments of their models' safety and performance.
Topics: AI Anthropic OpenAI Publishing Restrictions Regulation