Is Greynoise so AI-pilled that they didn't even notice they left the prompts in their brief preview? When your SVP is spending time recruiting on Twitter I guess that shit is to be expected. Yet another example of why I don't trust companies, including startups. Look how fast Greynoise went from legit research to a Nazi slop shop.

https://www.greynoise.io/resources/at-the-edge-clear-081726

@cR0w Maybe I just overestimate how much people care(or am failing to reckon with the possibility that leaving prompts in is vice-signaling to show how AI-centric you are); but the bit that always surprises me about leaving the user's prompts in is that whatever tool you are using should have a clear record of what those were and be easily able to ask if you want to remove them, prominently highlight them, etc. so they can be scrubbed with little effort.

@fuzzyfuzzyfungus That's what I would think as well but I'm not going to try to pretend to understand the thought process of the slop jockeys.

@cR0w It does seem like a cognitive risk to get too close. It just seems odd(vs. the much harder case of distinguishing the "Sure, here's the output you wanted in the style you wanted!" which is not distinct from the rest of the bot output).

Similar thing with totally fake citations vs. correct-but-misleading ones: you telling me that nobody has bolted on a search-engine-like-from-the-before-times step that can verify whether a citation exists?

@cR0w
Idk i'd read this as a suggestion to the reader. Probably still slop, but probably not the prompt.

@FritzAdalis Is that what it is? That's why I was asking. It's weird and doesn't make sense to me, a human.

@cR0w
I guess? Or maybe I'm an ai and don't realize it yet. That would suck.

@cR0w Is this what they mean by a GreyNoise prompt injection?

@FritzAdalis What if we all are?

@wdormann @cR0w are there screenshots of that?

@FritzAdalis @cR0w I’m looking at the last point (4) and it also reads like a suggestion to the readers. Based on the sentence structure, I’d bet it’s still AI slop, but perhaps indeed the output instead of the input.

@ClickyMcTicker @FritzAdalis That makes sense, which also explains why the sentence doesn't make sense.

@buherator @cR0w @wdormann well, im working on a talk called 'claude is your insider threat', you see - version 2! and its going to sweden the second week of sept. this version will have some more tips and tricks - and being able to show the audience how badly it looks when you fuck up to that magnitude would be a great visual indicator :D

@Viss @buherator @cR0w @wdormann I would very much like to attend your talk. Is it still tickets available, or live streamed somewhere?

@fere @buherator @cR0w @wdormann it's going to Sec-T in stockholm, I think the conference may be sold out, but I'm not sure - they do livestreams, I think. I'll know more later, but if they dont for some reason, the talks do end up on youtube. version 1 from securityfest is up there already. just search for "claude is your insider threat now"

@Viss @cR0w
Eh, it looks like I'm victim of playing the telephone game with my own memory. I guess it wasn't their own PoC. It was just a PoC.

From my notes in 2023:

=====
So my theories are:
1) OwnCloud recognized (or was told) that they have a phpinfo() page that can successfully be reached anonymously.
2) OwnCloud recognized that in docker installs, a successfully reached phpinfo() will leak the admin user/pass.
3) A CVSS 10 was assigned and advisory released, without actually confirming that docker installs are exploitable.
4) Somebody created a PoC exploit based on the public advisory, without confirming that it works.
5) Folks like Greynoise say that CVE-2023-49103 is being exploited in the wild because they saw traffic (from the above PoC).
6) Step 5) gets amplified in the usual infosec manner.
=====

Greynoise made a follow-up post admitting that the PoC didn't work

@cR0w @ClickyMcTicker @FritzAdalis ITT i learned:

• “slopjockeys”
• “vice-signaling”
• The unknowability of my own status as an AI or regular I.

@todb @ClickyMcTicker @FritzAdalis At least you have an I to be unsure about. 😅

@cR0w @ClickyMcTicker @FritzAdalis you’re picking up what I’m putting down. :)

I have spent way more time than usual in the company of eschatologists, lately.

@todb @cR0w @ClickyMcTicker @FritzAdalis whats 'vice-signaling'?

@cR0w I'm pretty sure Claude generates the reports, they aren't copy/pasted, but the whole thing definitely sounds AI generated

Am I crazy or is the headline "exploited vulnerabilities are exploited"?

@iagox86 @cR0w every org who has removed the human in the loop and just outputs all the work to claude - and i dont mean 'they built a tool using claude and the tool does the work' but literally its just a garden hose of work requests going into an anthropic api, and 'the work product coming out'... have built a house out of c4 and are having a taser fight in the livingroom

@iagox86 I think the idea is that orgs are getting exploited not because they don't know about the vulnerabilities but because they don't understand their attack surface exposure. But IDK, that might be a little too generous of an interpretation.