There's one very important thing I would like everyone to try to remember this week, and it is that AI companies are full of shit
-
@jenniferplusplus Open AI made similar claims about their model being so good it was dangerous and they weren't going to release it. In 2019. https://techcrunch.com/2019/02/17/openai-text-generator-dangerous/
@fancysandwiches oh wow, a headline that describes these things as text generators.
How far we've fallen
-
@dngrs @jenniferplusplus I'm sorry, I know thinking conceptually isn't easy for everyone, I tried using AlphaFold because some people have an easier time when presented with examples.
Why would there be an increase in CVEs? If I was an actor with nation-state levels of access to compute, why would I waste all that compute on zero days, only to then publish CVEs about them?
Even the most AI skeptic maintainers start to admit that LLMs are getting good at finding bugs. I understand cynicism is seen as cool nowadays but I think it's intellectually lazy
daniel:// stenberg:// (@bagder@mastodon.social)
I ran a quick git log grep just now. Over the last ~6 months or so, we have fixed over 200 bugs in #curl found with "AI tools".
Mastodon (mastodon.social)
@budududuroiu @dngrs you may as well stop, you're not going to convince me to trust them. Only anthropic can do that, because they have truly earned my distrust.
-
There's one very important thing I would like everyone to try to remember this week, and it is that AI companies are full of shit
Only rarely do their claims actually bear scrutiny, and those are only the mildest of claims they make.
So, anthropic is claiming that their new, secret, unreleased model is hyper competent at finding computer security vulnerabilities and they're *too scared* to release it into the wild.
Except all the AI companies have been making the same hypercompetence claims about literally every avenue of knowledge work for 3+ years, and it's literally never true. So please keep in mind the highly likely possibility that this is mostly or entirely bullshit marketing meant to distract you from the absolute garbage fire that is the code base of the poster child application for "agentically" developed software
You may now resume doom scrolling. Thank you
@jenniferplusplus Agree that it is mostly for marketing & investors.
But the article was technical enough, that I think there is an improvement here that no other model has. And if true, it would be great for vulnerability scanning/hardening in general (bad that attackers would have access to it though).
-
There's one very important thing I would like everyone to try to remember this week, and it is that AI companies are full of shit
Only rarely do their claims actually bear scrutiny, and those are only the mildest of claims they make.
So, anthropic is claiming that their new, secret, unreleased model is hyper competent at finding computer security vulnerabilities and they're *too scared* to release it into the wild.
Except all the AI companies have been making the same hypercompetence claims about literally every avenue of knowledge work for 3+ years, and it's literally never true. So please keep in mind the highly likely possibility that this is mostly or entirely bullshit marketing meant to distract you from the absolute garbage fire that is the code base of the poster child application for "agentically" developed software
You may now resume doom scrolling. Thank you
@jenniferplusplus Worth a follow for that post alone. Hi, I'm Bill.
-
There's one very important thing I would like everyone to try to remember this week, and it is that AI companies are full of shit
Only rarely do their claims actually bear scrutiny, and those are only the mildest of claims they make.
So, anthropic is claiming that their new, secret, unreleased model is hyper competent at finding computer security vulnerabilities and they're *too scared* to release it into the wild.
Except all the AI companies have been making the same hypercompetence claims about literally every avenue of knowledge work for 3+ years, and it's literally never true. So please keep in mind the highly likely possibility that this is mostly or entirely bullshit marketing meant to distract you from the absolute garbage fire that is the code base of the poster child application for "agentically" developed software
You may now resume doom scrolling. Thank you
@jenniferplusplus "our magic machine found a 30 year old security vulnerability!"
OK, what's the CVE link? These companies never show proof besides saying "it totally did the thing, you guyzzz plz giv moar billionz"
-
@jenniferplusplus I seriously doubt this is smoke and mirrors, recent models have improved significantly for cybersec and the industry is noticing:
daniel:// stenberg:// (@bagder@mastodon.social)
The challenge with AI in open source security has transitioned from an AI slop tsunami into more of a ... plain security report tsunami. Less slop but lots of reports. Many of them really good. I'm spending hours per day on this now. It's intense.
Mastodon (mastodon.social)
Linux kernel czar says AI bug reports aren't slop anymore
Interview: Greg Kroah-Hartman can't explain the inflection point, but it's not slowing down or going away
(www.theregister.com)
The industry consensus seems to be that there's going to be a torrent of vulnerabilities being found in all sorts of software, and they're not prepared to handle the blast radius. It's not surprising that Anthropic wants to give a select few a head start to tackle them. It would be nice if their token fund was open to all OSS projects to apply.
I'm also pressing "X doubt" that you spend months coordinating between AWS, Apple, Microsoft, Google, and the Linux Foundation to organise this just because your tool's code leaked online.
@budududuroiu @jenniferplusplus Let's talk about JavaScript. Have you ever looked at your browser's developer console? On any major website on the planet, there are 8 trillion errors in every one. Two-thirds of them are vulnerabilities, but none of them are exploitable or matter for anything at all. That is what is being found.
Those kinds of errors I've been reviewing, all the ones Daniel's been reviewing too, and I'm seeing it over and over. "Yes, okay, technically that is the buffer overrun, but it doesn't matter because you can't ever get to it!"
-
@jenniferplusplus It's also important that to whatever extent this product actually works (I'm as skeptical as you are), it fundamentally preferences the attacker. The product has way too many false positives to run in CI, so the defender can only use it as part of an occasional audit. The attacker doesn't care about CI or development friction, and wins by finding one exploit in an entire stack, even if they have to wade through many false positives to find it.
@jedbrown @jenniferplusplus The asymmetry is the core thing that concerns me. I can say that empirically starting somewhere last year LLM-assisted bug hunting started to be effective. The false positives are avoidable but the cost of remediation has not gone down with the cost of exploits. This new model may make the situation worse but we're already in it.
-
A couple people seem very invested in me being wrong about this assessment. All I can say is that this would be the first time I have misclassified an AI claim as bullshit
So here's the other thing that bothers me about all this. Regardless of the eventual results, this thing they're doing is *incredibly* resource intensive. They routinely spend billions of dollars on training these models, and billions more on operating them. It's not simple to parse out what fraction of that is directly attributable to the massive scale vuln finder/fabricator. But for the sake of argument lets just pick a plausible number, and call it 50-100 million dollars.
What could we have gotten for 50-100 million dollars of sponsorship for security audits? Prior to this, the largest single investment into FOSS security I'm aware of was the 2015 audit of openssl, after the heartbleed incident. It's hard to find precise costs for that, but I found a few sources estimating 1.2 million dollars, and that is arguably the most security critical piece of software in the world.
But suddenly there's 100x more resources available to do this work, now that producing the artifact can be done with stolen labor? Now that they can externalize the cost of false positives onto the already mostly unpaid maintainers of these projects? Even if their claims are true, which we have no reason to believe and very good reason not to, it's still a travesty
-
So here's the other thing that bothers me about all this. Regardless of the eventual results, this thing they're doing is *incredibly* resource intensive. They routinely spend billions of dollars on training these models, and billions more on operating them. It's not simple to parse out what fraction of that is directly attributable to the massive scale vuln finder/fabricator. But for the sake of argument lets just pick a plausible number, and call it 50-100 million dollars.
What could we have gotten for 50-100 million dollars of sponsorship for security audits? Prior to this, the largest single investment into FOSS security I'm aware of was the 2015 audit of openssl, after the heartbleed incident. It's hard to find precise costs for that, but I found a few sources estimating 1.2 million dollars, and that is arguably the most security critical piece of software in the world.
But suddenly there's 100x more resources available to do this work, now that producing the artifact can be done with stolen labor? Now that they can externalize the cost of false positives onto the already mostly unpaid maintainers of these projects? Even if their claims are true, which we have no reason to believe and very good reason not to, it's still a travesty
-
So here's the other thing that bothers me about all this. Regardless of the eventual results, this thing they're doing is *incredibly* resource intensive. They routinely spend billions of dollars on training these models, and billions more on operating them. It's not simple to parse out what fraction of that is directly attributable to the massive scale vuln finder/fabricator. But for the sake of argument lets just pick a plausible number, and call it 50-100 million dollars.
What could we have gotten for 50-100 million dollars of sponsorship for security audits? Prior to this, the largest single investment into FOSS security I'm aware of was the 2015 audit of openssl, after the heartbleed incident. It's hard to find precise costs for that, but I found a few sources estimating 1.2 million dollars, and that is arguably the most security critical piece of software in the world.
But suddenly there's 100x more resources available to do this work, now that producing the artifact can be done with stolen labor? Now that they can externalize the cost of false positives onto the already mostly unpaid maintainers of these projects? Even if their claims are true, which we have no reason to believe and very good reason not to, it's still a travesty
@jenniferplusplus 100 million dollars of sponsorship for FOSS project security audits doesn't sell a promise that soon all the humans can be fired.
-
So here's the other thing that bothers me about all this. Regardless of the eventual results, this thing they're doing is *incredibly* resource intensive. They routinely spend billions of dollars on training these models, and billions more on operating them. It's not simple to parse out what fraction of that is directly attributable to the massive scale vuln finder/fabricator. But for the sake of argument lets just pick a plausible number, and call it 50-100 million dollars.
What could we have gotten for 50-100 million dollars of sponsorship for security audits? Prior to this, the largest single investment into FOSS security I'm aware of was the 2015 audit of openssl, after the heartbleed incident. It's hard to find precise costs for that, but I found a few sources estimating 1.2 million dollars, and that is arguably the most security critical piece of software in the world.
But suddenly there's 100x more resources available to do this work, now that producing the artifact can be done with stolen labor? Now that they can externalize the cost of false positives onto the already mostly unpaid maintainers of these projects? Even if their claims are true, which we have no reason to believe and very good reason not to, it's still a travesty
@jenniferplusplus while I agree with the "AI companies are mostly full of shit" part, this would be the first kind of announcement like this I am taking semi-seriously.
Here's what's been happening the last couple of months, and this is with _current_ models. There are step functions at play, and I think the step function from "at least some skill needed to wield an LLM to find security issues" to "everybody with a $200 can exploit every OS/browser out there" should be considered very carefully.
Nicholas Carlini saying he found more bugs in 2 weeks than in his entire career with Mythos is not something I can dismiss.
Or daniel stenberg, certainly someone with actual authority and experience compared to me showing the current situation:
daniel:// stenberg:// (@bagder@mastodon.social)
I ran a quick git log grep just now. Over the last ~6 months or so, we have fixed over 200 bugs in #curl found with "AI tools".
Mastodon (mastodon.social)
daniel:// stenberg:// (@bagder@mastodon.social)
If your Open Source project sees a steep increase in number of high quality security reports (mostly done with AI) right now (#curl, Linux kernel, glibc confirmed) please tell me the name of this project. (I'd like to make a little list for my coming talk on this.)
Mastodon (mastodon.social)
-
So here's the other thing that bothers me about all this. Regardless of the eventual results, this thing they're doing is *incredibly* resource intensive. They routinely spend billions of dollars on training these models, and billions more on operating them. It's not simple to parse out what fraction of that is directly attributable to the massive scale vuln finder/fabricator. But for the sake of argument lets just pick a plausible number, and call it 50-100 million dollars.
What could we have gotten for 50-100 million dollars of sponsorship for security audits? Prior to this, the largest single investment into FOSS security I'm aware of was the 2015 audit of openssl, after the heartbleed incident. It's hard to find precise costs for that, but I found a few sources estimating 1.2 million dollars, and that is arguably the most security critical piece of software in the world.
But suddenly there's 100x more resources available to do this work, now that producing the artifact can be done with stolen labor? Now that they can externalize the cost of false positives onto the already mostly unpaid maintainers of these projects? Even if their claims are true, which we have no reason to believe and very good reason not to, it's still a travesty
@jenniferplusplus OpenSSL is important to the world. Software for which a CTO might be held responsible is important to that CTO. There should be more overlap, but there isn’t.
-
@jenniferplusplus while I agree with the "AI companies are mostly full of shit" part, this would be the first kind of announcement like this I am taking semi-seriously.
Here's what's been happening the last couple of months, and this is with _current_ models. There are step functions at play, and I think the step function from "at least some skill needed to wield an LLM to find security issues" to "everybody with a $200 can exploit every OS/browser out there" should be considered very carefully.
Nicholas Carlini saying he found more bugs in 2 weeks than in his entire career with Mythos is not something I can dismiss.
Or daniel stenberg, certainly someone with actual authority and experience compared to me showing the current situation:
daniel:// stenberg:// (@bagder@mastodon.social)
I ran a quick git log grep just now. Over the last ~6 months or so, we have fixed over 200 bugs in #curl found with "AI tools".
Mastodon (mastodon.social)
daniel:// stenberg:// (@bagder@mastodon.social)
If your Open Source project sees a steep increase in number of high quality security reports (mostly done with AI) right now (#curl, Linux kernel, glibc confirmed) please tell me the name of this project. (I'd like to make a little list for my coming talk on this.)
Mastodon (mastodon.social)
@mnl I'm not sure what I'm supposed to do with this. It feels like it's meant to dispute something I'm saying, but this is the same dynamic. The actual cost of operating these tools is 50-100x greater than the vendors are charging, which the vendors are doing in the hope that it eventually becomes an inextricable part of all work, completely eliminating labor as a social power.
Your hypothetical looks very different when it's "everybody with $20,000 (per month) can exploit every browser/os out there." Which is actually true now. It was true 6 months ago. It's been true for as long as we've had software that you could identify vulnerabilities in whatever software you wanted by paying a generous salary to full time researchers.
That's not what capital chose to do. And it bothers me that everyone is just adopting the capitalist framing on every goddamn word these companies spit out, as long as one of those words is AI
-
@mnl I'm not sure what I'm supposed to do with this. It feels like it's meant to dispute something I'm saying, but this is the same dynamic. The actual cost of operating these tools is 50-100x greater than the vendors are charging, which the vendors are doing in the hope that it eventually becomes an inextricable part of all work, completely eliminating labor as a social power.
Your hypothetical looks very different when it's "everybody with $20,000 (per month) can exploit every browser/os out there." Which is actually true now. It was true 6 months ago. It's been true for as long as we've had software that you could identify vulnerabilities in whatever software you wanted by paying a generous salary to full time researchers.
That's not what capital chose to do. And it bothers me that everyone is just adopting the capitalist framing on every goddamn word these companies spit out, as long as one of those words is AI
@jenniferplusplus I don't think i made a hypothetical? I don't disagree with the rest, but I wouldn't call this announcement bullshit.
I don't think saying that LLMs have gotten scaringly good at finding vulnerabilities (not hypothetical) is adopting the capitalist framing, in fact it's something that as a person supporting opensource and right to privacy, needs to be taken pretty seriously, since we can assume that these tools are in the hands of the government.
There's a fair amount of people (and yes, "AI companies") combining more traditional approaches to vulnerability finding with small models with known externalities to do similar work, one example I could find (I'm not a security's person) as a direct reaction to the mythos announcement: https://aisle.com/blog/ai-cybersecurity-after-mythos-the-jagged-frontier
-
@jenniferplusplus I don't think i made a hypothetical? I don't disagree with the rest, but I wouldn't call this announcement bullshit.
I don't think saying that LLMs have gotten scaringly good at finding vulnerabilities (not hypothetical) is adopting the capitalist framing, in fact it's something that as a person supporting opensource and right to privacy, needs to be taken pretty seriously, since we can assume that these tools are in the hands of the government.
There's a fair amount of people (and yes, "AI companies") combining more traditional approaches to vulnerability finding with small models with known externalities to do similar work, one example I could find (I'm not a security's person) as a direct reaction to the mythos announcement: https://aisle.com/blog/ai-cybersecurity-after-mythos-the-jagged-frontier
@mnl My point is that you're reading these things like a warning, where you should be reading them like a threat
-
@mnl My point is that you're reading these things like a warning, where you should be reading them like a threat
@jenniferplusplus a threat to? My livelihood as a programmer? The industry? I agree. But it is not an empty threat (meaning, I'm pretty sure this is real and that they are not just putting up such a disclosure announcement for hype and boost).
-
@jenniferplusplus a threat to? My livelihood as a programmer? The industry? I agree. But it is not an empty threat (meaning, I'm pretty sure this is real and that they are not just putting up such a disclosure announcement for hype and boost).
@jenniferplusplus this is maybe more what i'm reacting to. don't dismiss this stuff too quickly and bathe yourself in false comfort. If you are working on software, there's a reasonable chance these things can do a significant chunk of your job better than you. That they can't necessarily do it all, or do so for an extravagant amount of resources doesn't change that. I also don't want to sound contrarian, I know I might be a bit too autistic in my communication style (and I'm just as frustrated and anxious and exhausted like the rest of us).
-
@jenniferplusplus a threat to? My livelihood as a programmer? The industry? I agree. But it is not an empty threat (meaning, I'm pretty sure this is real and that they are not just putting up such a disclosure announcement for hype and boost).
@mnl when a mafia boss walks into a shop and talks about how much of a shame it would be if something happened to the place, that's also not an empty threat. That's the whole point. You can choose to pay them off, or not. What you absolutely do not do is run to all of your neighbors and redeliver the same threat
-
@mnl when a mafia boss walks into a shop and talks about how much of a shame it would be if something happened to the place, that's also not an empty threat. That's the whole point. You can choose to pay them off, or not. What you absolutely do not do is run to all of your neighbors and redeliver the same threat
@jenniferplusplus true, I hope that's not what I'm doing when I say "there's something to this and you need to pay attention to the impact of LLMs on security", even if I think anthropic is run by dangerous clowns (like you have mythos, and also your other stuff is maybe the most broken software I've ever used
) -
@jenniferplusplus this is maybe more what i'm reacting to. don't dismiss this stuff too quickly and bathe yourself in false comfort. If you are working on software, there's a reasonable chance these things can do a significant chunk of your job better than you. That they can't necessarily do it all, or do so for an extravagant amount of resources doesn't change that. I also don't want to sound contrarian, I know I might be a bit too autistic in my communication style (and I'm just as frustrated and anxious and exhausted like the rest of us).
@mnl @jenniferplusplus you seem fucking exhausting and have a long history on your public profile of AI boosterism so it’s not surprising that your response to both my and Jennifer’s posts is bland hype that doesn’t respond to any of the facts we’ve put forth
oh we’ll be left behind if we don’t adopt this terrible crap? good. leave us the fuck alone.
