Of the major web-based LLM frameworks in my roster (ChatGPT, Copilot, Claude, Gemini, and Perplexity), core functionality during the Cloudflare outage seems to be fine -- except for Claude, which is completely unusable.
If you're a "security researcher", and you send an email asking how to report vulnerabilities for a web property that I operate, and that web property has a security.txt ... you're getting marked as spam.
"If you're touching the brake or turning the steering wheel before activating the turn signal, you're doing it wrong" appears to be largely lost knowledge.