I largely agree with you, but I honestly understand their perspective. And I would even go as far to say that Daybreak blue is a bit TOO permissive, compared to what is actually proved. Imo, if OpenAI had the background checks like Anthropic do in strictness for their CVP program, then yes, make it this permissive. But for somewhere where the only thing you actually need is KYC, this is way too permissive. I shouldn’t be allowed to make a chromium exploit with only KYC. We already see how threat actors sell crypto KYC accounts for relatively cheap, and chances are they are probably also already selling access to Daybreak blue accounts as well. Anthropic on the other hand are a bit too strict on their program. I think you should at least be able to get a slightly less restrictive version for vulnerability research/discovery, since the default safeguards provide a kind of fake access to vulnerability research. And then tiers for active probing/attacking. Idk exactly the best model as I haven’t thought about it enough for it, yet. But that’s my opinions.