Lack of standards for testing makes results of safe harbor testing unusable. Safety isn’t black and white

Red teaming is a critical tool for improving the safety and security of AI systems. It involves adversarially testing a technological system to identify potential vulnerabilities. Today, researchers and AI developers employ a wide range of red teaming techniques to test their AI systems, each with its own advantages and disadvantages. The lack of standardized practices for AI red teaming further complicates the situation. Developers might use different techniques to assess the same type of threat model, and even when they use the same technique, the way they go about red teaming might look quite different in practice. This inconsistency makes it challenging to objectively compare the relative safety of different AI systems.

Insert theory is desired

CP links to the net benefit(s) because ________________________________________