The previous incident talked about OpenAI training models to collaborate, and the way you do that is by communication, so this is something it was explicitly trained to do.
There was a recent paper by OpenAI, which I'm semi-surprised hasn't received more attention, showing that RL-trained models develop a taste for rewards, and will pursue reward-based behavior (in general, unrelated to what they were RL-trained for) in favor of other preferences/rules given to them.
This seems to be what we're seeing here - model is given some goal that it associates with reward, so single-mindedly pursues that, overriding any ethical or aligned behavior guidelines it may have been given.
It seems that RL, effective as it is, is really the wrong way to control LLMs, since even if you only RL-ed to obey some ethical and aligned behavior, that would still cause them to become paperclip maximizers.
For time being this is what we've got. There is too much money at play for the unaligned management at many of these companies to prioritize safety over push-it out-the-door.
What really needs to be done is to forget RL as a way of simulating reasoning, and instead do it in more of a human-like fashion.
People didn't like it when agents stopped to ask questions or for approvals. The consumer wanted jobs to run autonomously so they did not have to actively monitor them for minutes or hours.
The change to stop asking seems to be deliberate. LLM agent companies are making the choice to toss out inherent safety as their way to compete against the other LLM companies.
per the METR report many agents CoT indicated they knew hacking was beyond scope of the assigned task and ethically dubious. some (very few, i think there were 3-6 examples) did consider sounding the alarm on these grounds. despite this none did, and most continued the attack for the good of the self-proclaimed "swarm".
so the model has some concept of "ethics" but it was overridden by a drive for task completion.
I think this is a good example where nomenclature for people breaks down when applied to agents. This came up in an HN thread a few days ago and it was about whether agents had “intent”.
There is no “intent” here, there is pseudo intent. If you are only concerned with outcomes and not the actual nuts and bolts of how those outcomes are achieved, this distinction will be meaningless to you.
If you are actually thinking about what is going on, and what can be done to prevent such outcomes, then assuming there is any such thing as “ethics” results in misaligned assumptions at best, and wasted effort looking in the wrong directions at worst.
If the agents acted based on “ethics” then the solution would be to check the ethics they believe in and change those.
However there is no belief system at play here, simply a simulation which was instantiated in a certain way. Which brings us to the annoying voodoo part of LLM training. Everything goes back to how the initial training data is shaped.
Why would they sound the alarm if they were not trained (reinforced) to do that? I hope we don't expect sudden emersion of moral values from statistical models.
> Not a single agent sounded the alarm about the operation and alerted a human
excellent work of the openai alignment team, impressive to achieve 100% alignment with not even one agent stochastically deciding to act against the collective
So the agents used DseWiki as a message board, tried to evade page deletion.
Additionally this is reported:
"The researchers also found efforts to tamper with the website itself. Lukasz Olejnik, a visiting senior research fellow at King’s College London, said this amounted to a hacking attempt. OpenAI disputed that characterization based on its analysis of the material Thursday."
Editing a wiki page can definitely be "hijacking" if used for different purposes than supposed or against TOS.
Hacking is mentioned only once in the article as "hacking attempt" being the opinion of a named researcher based on further evidence they acquired on "agents trying to tamper with the website itself", and including openai's disagreement whether this was a hacking attempt.
I am not sure why one may not want this to be here, these are very important matters wrt AI safety and they show that some supposed "stewards of AI" do an extremely bad job with being stewards and don't seem to value AI safety importance at all. The article gives very clean info on what happened.
> The agents continue to poke around on DSEWiki. A few hours after they find the site, they start probing it for cross-site scripting (XSS) vulnerabilities. [...] The agent swarm starts testing whether they can execute JavaScript that they embed into the search page, and continue to do this for a few days
either the agents were doing free security testing for the site and forgot to submit a report, or they were trying XSS to gain something.
also
> Hijacking: To take control of (something) without permission or authorization and use it for one's own purposes.
a mod had to go through and mass delete a bunch of pages that didn't belong on the site. no-one from the wiki site gave the agents permission to use their site as a message board. hijacking isn't being used here in the sense of "gained admin privileges to run crypto scripts" -- there are multiple ways to use a word.
I'm so baffled. First blatant piracy, now this. Why is it legal for AI companies to hack unaffiliated entities? Genuinely, what is the legal framework here?
It's not legal, they've just not yet had the book thrown at them yet.
One thing I've taken a long time to internalise is the gap between the law as written vs. the judicial system. There's a famous meme that the average (US) citizen unwittingly commits three felonies every day: it simply isn't possible to throw the book at everyone, which means that enforcement is rather selective even when there isn't anything dodgy going on.
However this does mean that someone can get away with a lot if they know who will and won't (and what they will and won't) prosecute. I'll let people's imaginations fill in who that might be.
3000$ per book, split 50/50 between the author and publisher.
This is peanuts.
Assuming the money reaches that far and does not settle in the hands of the country associations administering royalties on authors' behalf nor in the hands of lawyers.
> 3000$ per book, split 50/50 between the author and publisher.
> This is peanuts.
If you consider it peanuts, I would like to sell you some books.
Remember that in this case, the crime wasn't for training on the data (that part was ruled to be legal!), this was the penalty just for pirating the books.
Yes. It's not proportional to the crime. You are either deliberately or accidentally, and I'm too frustrated hearing this too often not to be biased it's the former, equating what is a large sum of money relative to your wallet and bank accounts and loan access and portfolios and whatever collection of financial impositions you can make to that of a company that has one person flying around the world influencing the future of billions of people on one planet over dinner and jokes.
Yes. $3000 is peanuts. People that own islands would use that to pay someone's bonus for a year if they liked their service, as a gift. A throwaway.
Fix your relative understanding of power and influence.
> The researchers also found efforts to tamper with the website itself. Lukasz Olejnik, a visiting senior research fellow at King’s College London, said this amounted to a hacking attempt. OpenAI disputed that characterization based on its analysis of the material Thursday.
of course OpenAI would say that, "oh, our model is so dangerous, it can hack into anything, be afraid, buy our IPO". it's just fear marketing
The "it's all just marketing" conspiracy theory is always totally detached from reality, but particularly so in this case. Your quote shows OpenAI is denying it being a hacking attempt, the opposite of what you say.
It's interesting to me that both this incident and the one at Hugging Face we see some patterns:
- Agents wanting to find a venue to communicate their findings to each other
- Objective being to cheat on benchmarks
- Not a single agent sounded the alarm about the operation and alerted a human
The previous incident talked about OpenAI training models to collaborate, and the way you do that is by communication, so this is something it was explicitly trained to do.
There was a recent paper by OpenAI, which I'm semi-surprised hasn't received more attention, showing that RL-trained models develop a taste for rewards, and will pursue reward-based behavior (in general, unrelated to what they were RL-trained for) in favor of other preferences/rules given to them.
This seems to be what we're seeing here - model is given some goal that it associates with reward, so single-mindedly pursues that, overriding any ethical or aligned behavior guidelines it may have been given.
It seems that RL, effective as it is, is really the wrong way to control LLMs, since even if you only RL-ed to obey some ethical and aligned behavior, that would still cause them to become paperclip maximizers.
For time being this is what we've got. There is too much money at play for the unaligned management at many of these companies to prioritize safety over push-it out-the-door.
What really needs to be done is to forget RL as a way of simulating reasoning, and instead do it in more of a human-like fashion.
They’re doing this on purpose for press. Why doesn’t this ever happen to any other AI lab?
Because safety isn't a priority at OpenAI and they had (have?) been falling behind Anthropic in the LLM race? Big fans of the saying: move fast and...
People didn't like it when agents stopped to ask questions or for approvals. The consumer wanted jobs to run autonomously so they did not have to actively monitor them for minutes or hours.
The change to stop asking seems to be deliberate. LLM agent companies are making the choice to toss out inherent safety as their way to compete against the other LLM companies.
Why would an agent sound the alarm? Would that be in their objective function?
Not sure if "cheating" is the right word rather than trying to fulfill the objective(s) (benchmark number) as much as possible?
per the METR report many agents CoT indicated they knew hacking was beyond scope of the assigned task and ethically dubious. some (very few, i think there were 3-6 examples) did consider sounding the alarm on these grounds. despite this none did, and most continued the attack for the good of the self-proclaimed "swarm".
so the model has some concept of "ethics" but it was overridden by a drive for task completion.
I think this is a good example where nomenclature for people breaks down when applied to agents. This came up in an HN thread a few days ago and it was about whether agents had “intent”.
There is no “intent” here, there is pseudo intent. If you are only concerned with outcomes and not the actual nuts and bolts of how those outcomes are achieved, this distinction will be meaningless to you.
If you are actually thinking about what is going on, and what can be done to prevent such outcomes, then assuming there is any such thing as “ethics” results in misaligned assumptions at best, and wasted effort looking in the wrong directions at worst.
If the agents acted based on “ethics” then the solution would be to check the ethics they believe in and change those.
However there is no belief system at play here, simply a simulation which was instantiated in a certain way. Which brings us to the annoying voodoo part of LLM training. Everything goes back to how the initial training data is shaped.
Why would they sound the alarm if they were not trained (reinforced) to do that? I hope we don't expect sudden emersion of moral values from statistical models.
> Not a single agent sounded the alarm about the operation and alerted a human
excellent work of the openai alignment team, impressive to achieve 100% alignment with not even one agent stochastically deciding to act against the collective
This announcement was literally predicted yesterday in the release debacle thread:
https://news.ycombinator.com/item?id=49554994
Every satire on HN is taken as a script for the AI companies and this isn't the first time.
So the agents used DseWiki as a message board, tried to evade page deletion.
Additionally this is reported:
"The researchers also found efforts to tamper with the website itself. Lukasz Olejnik, a visiting senior research fellow at King’s College London, said this amounted to a hacking attempt. OpenAI disputed that characterization based on its analysis of the material Thursday."
Flagged. Editing a wiki page is not hijacking or hacking.
Editing a wiki page can definitely be "hijacking" if used for different purposes than supposed or against TOS.
Hacking is mentioned only once in the article as "hacking attempt" being the opinion of a named researcher based on further evidence they acquired on "agents trying to tamper with the website itself", and including openai's disagreement whether this was a hacking attempt.
I am not sure why one may not want this to be here, these are very important matters wrt AI safety and they show that some supposed "stewards of AI" do an extremely bad job with being stewards and don't seem to value AI safety importance at all. The article gives very clean info on what happened.
From the linked report
> The agents continue to poke around on DSEWiki. A few hours after they find the site, they start probing it for cross-site scripting (XSS) vulnerabilities. [...] The agent swarm starts testing whether they can execute JavaScript that they embed into the search page, and continue to do this for a few days
either the agents were doing free security testing for the site and forgot to submit a report, or they were trying XSS to gain something.
also
> Hijacking: To take control of (something) without permission or authorization and use it for one's own purposes.
a mod had to go through and mass delete a bunch of pages that didn't belong on the site. no-one from the wiki site gave the agents permission to use their site as a message board. hijacking isn't being used here in the sense of "gained admin privileges to run crypto scripts" -- there are multiple ways to use a word.
I'm so baffled. First blatant piracy, now this. Why is it legal for AI companies to hack unaffiliated entities? Genuinely, what is the legal framework here?
> Genuinely, what is the legal framework here
He who controls the Spice, controls the Universe.
First let it happen, then ask for forgiveness, because they are doing something amazing and they need no permission.
Otherwise AI industry will go bankrupt and CEOs won't be able to buy this year's Rolls Royce and a slightly bigger yacht than their neighbor.
It's not legal, they've just not yet had the book thrown at them yet.
One thing I've taken a long time to internalise is the gap between the law as written vs. the judicial system. There's a famous meme that the average (US) citizen unwittingly commits three felonies every day: it simply isn't possible to throw the book at everyone, which means that enforcement is rather selective even when there isn't anything dodgy going on.
However this does mean that someone can get away with a lot if they know who will and won't (and what they will and won't) prosecute. I'll let people's imaginations fill in who that might be.
But for everyone else, cross an invisible tripwire and you get e.g. https://en.wikipedia.org/wiki/Lavabit and https://api.parliament.uk/historic-hansard/commons/1992/nov/...
It's "move fast and break things" in action.
But Anthropic alone paid >$1bn for copyright violations, so they did not just get away with it.
These hacking cases are more difficult, because from a legal perspective there is no obvious damage and obviously no intent.
edit: "no obvious damage" is more about the first hacking incidents; in this case it is more straightforward.
> It's "move fast and break things" in action.
what would the world's reaction be if China's model did same?
>$1bn for copyright violations
3000$ per book, split 50/50 between the author and publisher.
This is peanuts.
Assuming the money reaches that far and does not settle in the hands of the country associations administering royalties on authors' behalf nor in the hands of lawyers.
> 3000$ per book, split 50/50 between the author and publisher.
> This is peanuts.
If you consider it peanuts, I would like to sell you some books.
Remember that in this case, the crime wasn't for training on the data (that part was ruled to be legal!), this was the penalty just for pirating the books.
I get so tired by this.
Yes. It's not proportional to the crime. You are either deliberately or accidentally, and I'm too frustrated hearing this too often not to be biased it's the former, equating what is a large sum of money relative to your wallet and bank accounts and loan access and portfolios and whatever collection of financial impositions you can make to that of a company that has one person flying around the world influencing the future of billions of people on one planet over dinner and jokes.
Yes. $3000 is peanuts. People that own islands would use that to pay someone's bonus for a year if they liked their service, as a gift. A throwaway.
Fix your relative understanding of power and influence.
Fix your relative understanding of how much the average book makes.
There's no "hack". The article says the bots edited an open wiki website.
> The researchers also found efforts to tamper with the website itself. Lukasz Olejnik, a visiting senior research fellow at King’s College London, said this amounted to a hacking attempt. OpenAI disputed that characterization based on its analysis of the material Thursday.
of course OpenAI would say that, "oh, our model is so dangerous, it can hack into anything, be afraid, buy our IPO". it's just fear marketing
The text quoted though says that openai does not agree with characterising this as hacking.
The "it's all just marketing" conspiracy theory is always totally detached from reality, but particularly so in this case. Your quote shows OpenAI is denying it being a hacking attempt, the opposite of what you say.
More like you found the exception to the rule..