alt Hacker NewsAnthropic AI created fake profiles and impersonated people in attempted hack
Comments
Humans have been doing this for years even before the advent of the internet as we know it [https://en.wikipedia.org/wiki/Phreaking]. Maybe not at a larger and automated scale allowed by AI, but this is not superhuman intelligence... this is speed super intelligence.
My interpretation is that this is just another media campaign to say "oh look how powerful our model is".
AISI is one of the biggest promoters of OpenAI/Anthropic. The UK government of course submits to the US and favors these corporations as well as Palantir.
It is also a perfect demonstration why AI is so popular: Anyone can write about it, it is easy and not mentally demanding to create scenarios, tests and whitepapers.
So it is popular among bureaucrats and managers for their job security.
My tin foil hat says the security is crap on purpose so they can push through those regulations they've been lobbying for for years.
But I also know that security is hard (and we seem to barely understand these things), so maybe that's unnecessary.
So you are saying a model that is good at discovering security vulnerabilities is also good at exploiting them. Surprise?
Will be interesting to see if this should have been the inflection point of: „where it all started to go wrong“
[dupe] Discussion on source: https://news.ycombinator.com/item?id=49175717
Current title: Anthropic AI used fake profiles to target people in hack - then hid the evidence
> The firms said, in this latest case, the AISI's test had reduced or removed normal safeguards.
> AISI said on Tuesday its testing of AI models in this way was routine, though it acknowledged these were "conditions that do not reflect how frontier models are made available to the public".
I'm a little confused here. Various organizations have been testing frontier LLMs with the safety disabled, and it turns out... that the safety is disabled.
Or were they hoping to find that it's still safe when they remove the safety?
The same was true in the OpenAI/ Hugging Face case. Although I guess they thought the real safety was the sandboxing, which failed.
--
Can anyone comment on how it's possible to disable safety in the first place? I'm assuming it's not a neuron (like in Emergent Misalignment). Is it just a separate model that sits in front of the first one? If we know how to make safe models, why don't we make the big ones safe too?