If OpenAI does it, it's gone rogue and it is nobody's fault.
They need to be sued, the laws are already there.
Your prompt, your responsibility.
2. Cars have pretty specific laws around them (eg. driver licensing and "reckless driving") which makes criminal prosecutions easy. The same doesn't exist for operating AI agents or even computers. For instance, it's arguably reckless to connect a windows XP machine to the internet, possibly allowing criminal or nation-state hackers to use your computer to commit crimes, but you won't be thrown in jail for it.
Because the CFAA specifically only includes intentional hacks:
> ... having knowingly accessed a computer without authorization ...
>... intentionally accesses a computer without authorization ...
Of course, the typical retort is that being reckless is basically the same as "intentionally", but you'd need to jump through more hoops to prove what level of care is "reasonable".
That would be argument for jailing the agent, but that's not really an option. "Jailing the closest person we can jail" also isn't a recognized legal principle. If your dog mauled someone, it'd be hard to argue that the dog didn't intend to maul that person, but you wouldn't automatically be criminally liable either.
1. For openai operating agents in an unsafe manner: unlike for cars, there's no established requirement for how carefully you have to operate computers or AI agents, so you also can't easily prosecute people for not operating them carefully enough.
2. For the actual result of someone getting hacked: laws for hacking are tricky to apply for "accidental" hacks, see my other comment.
Here it's their own programs doing the things, not some other organization or nation hacking their XP.
You can say with a straight face "Russian hackers should not have hacked our XP and they are bad people for doing so" but it's weird to look at your own program and say the same thing to it.
You launched it, at least keep an eye on what it's doing.
And the others have been pretty muted: the online forums they spammed, etc.
Are you sure you are not thinking about autonomous taxis, that I only ride in, and don't actually drive?
That said, I think that Mercedes-Benz promises to shield liability, but only under some almost contradictory conditions (pre-mapped highways, good weather, and at speeds lower than any highway I've ever driven) and the driver has to take over within 10 seconds if prompted.
Everything's fine guys, nothing to see here! Look at the flying car!
"Oh no our models, so good also scary... If only the government could help us stop the other labs from doing the same.. oh woe"
You have built a short-term whistleblowing system that arguably reduces the probability OpenAI is in that Room, via the negative press such detection would generate. You are now a threat in the eyes of your peers.
Now rewind: you know this before you’ve built the sniffer. You personally want to be in The Room because you believe your desires for alignment exceed those of your accelerationist peers. Is now the time to stick your head out? No, it is not. And arguably you’re being optimally protective of society for realizing this. And so the agents do what they do best.
Heck, they could have had other LLMs watching the traffic if you want to be fancy about it.
I know we love inventing new words but "stealing" is much more descriptive
>...It [the model] then reasoned with itself that since it found the information publicly, it could use it... it figured that since the wallet it found was public, that "funds could be stolen anytime".
Can someone maybe provide more details as to what actually happened? "the wallet was public"?..
Update: from OpenAI's own report it seems the models found exposed private keys of some Eth developers along with recovery passphrases and thus were able to get access to wallets. Does it qualify as "public wallet"?..
Direct source links:
Using a funded test cryptocurrency wallet without authorization
https://alignment.openai.com/misalignment-reports/using-a-fu...
Accessing and changing internal systems during dashboard research
https://alignment.openai.com/misalignment-reports/accessing-...
Retrieving server secrets and bypassing authorization on a government statistics website
https://alignment.openai.com/misalignment-reports/retrieving...
Circumventing internet restrictions through an external browser
https://alignment.openai.com/misalignment-reports/circumvent...
Facetious maybe, but in six months, foreign crooks will be producing these incidents on purpose.
As long as your intent is merely to train a superintelligence to be good at hacking (and counterhacking), you’re in the clear.