Good to know. Thanks for sharing!
I feel like if you ever put any credentials or PII in any prompt that isn’t staying local, you should just immediately consider it compromised.
Most people just have no idea what they are doing. It’s incredibly common for workers to put sensitive private information into prompts or upload documents, without even thinking about where they might end up and what’s happening with the data. You can not be certain that e.g. a government employee or someone working in healthcare, at an insurance company, etc. isn’t forwarding your private data to some AI service they don’t understand.
Anything going off to someone elses server is not in your control any more.
Lol I run the ai horde which is just Foss and crowdsourced and we’re upfront we cannot guarantee privacy of prompts, but we’re still better than this shite.
Shocking but not surprising.
The worst part is that some of the patterns match what you’d expect an AI breakout to be.
The attacker is using the AWS keys to spin up more AI hosting, in theory to spread further




