The 1st amendment, if that's what Anthropic claims, only goes so far.
Releasing Mythos or Fable is much like giving away Javelin missiles for free to everyone and anyone, but then crooks and hackers are part of that group.
Severe damages will result, even losses of life from 2nd and 3rd hand kinetic events, not missile explosions. Because of this danger I reluctantly favor government teams scrutinizing them and issuing approval before models are released, sorry.
Anthropic guys step over the line, they do need to be arrested to show them you just don't behave like this...
It's really just one or 2 steps away from a "ready to hack" exploit.
OMG, it already happened: As reported, within hours of Fable 5's public launch, a researcher known online as Pliny the Liberator posted on social media claiming to have "liberated" the model. The method required no code exploitation, no reverse engineering, no software vulnerabilities — just carefully crafted prompts.
The documented techniques included:
Unicode and Cyrillic character substitution, well, there's the tagline, cyrillic substitution indicates the Russky's already have it. More drain of millions of wealth from corporations into hacker accounts via crypto.
In the technique example, it sounds like they used a homograph (or homoglyph) method to trick the AI -- almost as if the AI has 'eyes' and reads like a human would. This is so interesting.
You think that if someone can get a model to write a beginner's guide to exploiting code that requires writing your own purposefully vulnerable program, then the creators of that model should be arrested?
The 1st amendment, if that's what Anthropic claims, only goes so far.
Releasing Mythos or Fable is much like giving away Javelin missiles for free to everyone and anyone, but then crooks and hackers are part of that group.
Severe damages will result, even losses of life from 2nd and 3rd hand kinetic events, not missile explosions. Because of this danger I reluctantly favor government teams scrutinizing them and issuing approval before models are released, sorry.
Anthropic guys step over the line, they do need to be arrested to show them you just don't behave like this...
It's really just one or 2 steps away from a "ready to hack" exploit.
OMG, it already happened: As reported, within hours of Fable 5's public launch, a researcher known online as Pliny the Liberator posted on social media claiming to have "liberated" the model. The method required no code exploitation, no reverse engineering, no software vulnerabilities — just carefully crafted prompts.
The documented techniques included:
In the technique example, it sounds like they used a homograph (or homoglyph) method to trick the AI -- almost as if the AI has 'eyes' and reads like a human would. This is so interesting.
You think that if someone can get a model to write a beginner's guide to exploiting code that requires writing your own purposefully vulnerable program, then the creators of that model should be arrested?
Why not? If you can get arrested for saying "free palestine" then why not for teaching computer hacking?
> According to a Wall Street Journal (WSJ) report, the technique in question was tested by researchers at Amazon,
Why not link that? Is this a dupe?
Found the primary source:
https://www.wsj.com/tech/ai/amazon-ceos-talks-with-u-s-offic...
Being discussed here: https://news.ycombinator.com/item?id=48519092
We should probably ban all of those Amazon models that can also be jailbroken.