pull down to refresh

33 sats \ 5 replies \ @optimism 14 Nov \ parent \ on: Researchers question Anthropic claim that AI-assisted attack was 90% autonomous AI

But if you send code to analyze to Claude, then Anthropic has that code. So they have the 0day code. You cannot ask Claude to analyze something without sending everything to anthropic.

write
preview
190 sats \ 4 replies \ @0xbitcoiner OP 14 Nov

Right. Maybe I missed it, but I didn’t see anywhere in the report saying the AI was the one that found the 0day vulnerability.

reply
55 sats \ 3 replies \ @optimism 14 Nov

Yeah they just imply it with the 90%. "Give SuperClaude a list of targets and yeah... pwned" lol

reply
203 sats \ 2 replies \ @0xbitcoiner OP 14 Nov

Correction: 80%-90% 🤠

Overall, the threat actor was able to use AI to perform 80-90% of the campaign, with human intervention required only sporadically (perhaps 4-6 critical decision points per hacking campaign).

https://www.anthropic.com/news/disrupting-AI-espionage
reply
33 sats \ 1 reply \ @optimism 14 Nov

Okay. Here's a mindfuck back at them: How do they have any idea what the humans did unless they were employing them?

reply
100 sats \ 0 replies \ @0xbitcoiner OP 14 Nov

¯\(ツ)/¯

reply