@dangoodin@infosec.exchange I'm experienced but not expert in this area, and it seems to me we are missing the "threat model" classification. We generally categorize somebody willing to spend $100,000 to break your messages as a "state-level actor", or at least Fortune 500 corporate actor or oligarch. You need a whole data center full of GPUs to iterate on these problems, and of course Anthropic has access to that. It's not "LLM" or "AI" that did this, it's raw compute power, and the willingness to spend it.