Is this in a lab, yes. Is this also what operators would run not in a lab, also yes. Everything in memory, done with off-the-shelf malware with a Rigging integration from run in an "agentic" loop. Sometimes the models do well, sometimes they do poorly, but progress is up and to the right...and the data we get is 🔥
For defenders, it does you no good to see an AI thing and immediately say "that's not possible". It's exactly what Proofpoint said when we reported Proofpudding...It's just here now.
GPUs are about speed and scale, that's what you'll get with AI driven ops, speed and scale. It doesn't undo all the defenses built up in networks.