I'm dubious of such a claim, both because it strikes me as unlikely but moreso because if it happened I would expect them to clamp down and not allow that information to get out. The only way it makes sense for them to allow the press to cover that is if it's some sort of propaganda or psyop flex of "Oh noes, look out China, our AI company built an AI so good that it broke into the NSA super fast. Better watch out."
Social engineering remains the most effective hack and that isn't going to change anytime soon.
I know a guy who used to weasel his way into government spaces he really had no business being in and almost all of it was done via social engineering. Like...absurd levels of physical access to classified environments. People really have no idea how fragile the human element of security truly is.
The term hack is thrown around a lot when 90% of the time it’s boiled down to incompetence. I can guarantee you someone on the NSA side was using an AI chat on a secret clearance computer and it was connected to the internet (which both is a No No.)
This. People think the government is some mythical all-powerful entity. In reality the government hires from the bottom of the barrell. The lowest of the low. The lower dregs of society. Smart, wise, competent, capable, successful people don't work in government. Only the failures in life, the losers, the weak, the stupid, the incompetent, and the pathologically toxic people seek employment in government because they can't hold down a job in the private sector.
It's a little of both. Anthropic and OpenAI both have a history of lying about the capabilities of their models to drum up hype. Anthropic in particular this time around has achieved some kind of regulatory capture by directly scaring government officials.
However, there is also real danger and capability here and dismissing it as a "chat bot" is uninformed. AI is writing math papers now. Recently Paul Erdős' unit distance conjecture was disproven by an AI model. And you might say, oh well most of a math proof is just trying a bunch of shit to see what works - well guess what, most of hacking is like that too.
Even the free claude model is capable of byte level hacking, I've used it myself.
I'm dubious of such a claim, both because it strikes me as unlikely but moreso because if it happened I would expect them to clamp down and not allow that information to get out. The only way it makes sense for them to allow the press to cover that is if it's some sort of propaganda or psyop flex of "Oh noes, look out China, our AI company built an AI so good that it broke into the NSA super fast. Better watch out."
I bet some saar at the NSA redeemed a phish.
Social engineering remains the most effective hack and that isn't going to change anytime soon.
I know a guy who used to weasel his way into government spaces he really had no business being in and almost all of it was done via social engineering. Like...absurd levels of physical access to classified environments. People really have no idea how fragile the human element of security truly is.
The term hack is thrown around a lot when 90% of the time it’s boiled down to incompetence. I can guarantee you someone on the NSA side was using an AI chat on a secret clearance computer and it was connected to the internet (which both is a No No.)
This. People think the government is some mythical all-powerful entity. In reality the government hires from the bottom of the barrell. The lowest of the low. The lower dregs of society. Smart, wise, competent, capable, successful people don't work in government. Only the failures in life, the losers, the weak, the stupid, the incompetent, and the pathologically toxic people seek employment in government because they can't hold down a job in the private sector.
Could be that NSA did update, and got hit by any number of packagmanager infected packages, like on NPM or AUR.
Advertising
It's a little of both. Anthropic and OpenAI both have a history of lying about the capabilities of their models to drum up hype. Anthropic in particular this time around has achieved some kind of regulatory capture by directly scaring government officials.
However, there is also real danger and capability here and dismissing it as a "chat bot" is uninformed. AI is writing math papers now. Recently Paul Erdős' unit distance conjecture was disproven by an AI model. And you might say, oh well most of a math proof is just trying a bunch of shit to see what works - well guess what, most of hacking is like that too.
Even the free claude model is capable of byte level hacking, I've used it myself.
Not seeing this mentioned anywhere but Twitter
economist