Comment by JohnMakin
Even if you were unwilling to change this behavior on the application layer or server side, you could add a directive in the proxy to prevent such large payloads from being accepted as an immediate mitigation step, unless they seriously need that parameter to have unlimited number of urls in it (guessing they have it set to some default like 2mb and it will break at some limit, but I am afraid to play with this too much). Somehow I doubt they need that? I don't know though.
Cloudflare is proxy in front of the API endpoint. After it became apparent that BugCrowd is tarpitting me and OpenAI didn't care to respond, I reported to Cloudflare via their bug bounty because I thought it's such a famous customer they'd forward the information.
But yeah, cloudflare did not forward the vulnerability to openai or prevent these large requests at all.