The onslaught includes LLMs finding bogus vulnerabilities and code that won't compile.
See full article...
See full article...
In an interview, Stenberg said that the reporter, Joshua Rogers, mostly used AI-powered code analyzer called ZeroPath.
“A clever person using a powerful tool,” Stenberg wrote.
But think of all the productivity you've gained!The onslaught includes LLMs finding bogus vulnerabilities and code that won’t compile.
Don't think that really applies here. This is people downloading the source code for cURL and then running AI agents on that. Then they get the result from it and blindly stick it in a reporting form to try to win a bounty.My people. You who have been wronged and beset by the horrors of an expanding slopverse…arise and unleash Hell!
Tarpits for AI (Ars Article)
After the bug reporter complained and reiterated the risk posed by the non-existent vulnerability, Stenberg jumped in and wrote: “You were fooled by an AI into believing that. In what way did we not meet our end of the deal?
I was more referring to the general strategy of going on the offensive against the slopverse. I like defensive measures as much as anyone, but I’m increasingly of the mind that the well must be poisoned as close to the source of this behavior as possible.Don't think that really applies here. This is people downloading the source code for cURL and then running AI agents on that. Then they get the result from it and blindly stick it in a reporting form to try to win a bounty.
From the article:
The people who just ask blindly put code into an AI code scanner will also be the same people who ask the AI to create a proof of concept attack and then never test it.What I didn't understand about the reporting was why they couldn't automatically gatekeep reports by requiring a proof-of-concept attack. Build a working exploit and then a human can review the details.
If you aggregated the impact of LLMs, I would bet heavily that the net effect on total human productivity is negative. Certainly there are positive areas, unfortunately drowned in the time suck that is 'slop'.But think of all the productivity you've gained!
They'd just get AI slop proof of concept attacks.What I didn't understand about the reporting was why they couldn't automatically gatekeep reports by requiring a proof-of-concept attack. Build a working exploit and then a human can review the details.
Think of all the power company workers employed, and their poor children if moronic AI scrapers get stopped. /SBut think of all the productivity you've gained!
Maybe a $5 submission fee could prevent these zero effort reports.
$5 might be too little considering how time consuming a bug report review can be. Make it $50. But then cURL team has to start doing taxes and whatnot
Just picked up Snow Crash on my Everand (non Amazon) reader.Something, something Torment Nexus….![]()
Or use AI to filter out the bad reports? Maybe with a set of whitelisted people who instantly go to people, since you trust them.
Sure that could become an issue too, but if you can get an AI group to donate time it'll drop their taxes a bit for the resources they donate to cURL.
Or add a fee for all submissions. That was a solution to email SPAM that I read about a while back. If we added a small fee for every email sent (pennies or fractions of a penny) then SPAM'ing people by the millions isn't profitable anymore. Tune the values to not hurt valid uses (not my forte), but block blindly bothering other people.
No, those bug reports typically include a "proof-of-concept" attack, but the attack itself is also AI-slop. Just watch this video and you'll see exactly why your idea doesn't work:What I didn't understand about the reporting was why they couldn't automatically gatekeep reports by requiring a proof-of-concept attack. Build a working exploit and then a human can review the details.
Or a more crude way to put it... if you keep shitting on my lawn I'm not hiring someone to go around cleaning it up, I'm just putting up a fence.Because it's a whole lot easier, and probably safer, to just not take bug reports. Why should the dev team spend it's time cleaning up someone else's mess?
It’s 100% profit for them to “spray and pray.” It requires zero effort or time from them and costs everyone else dearly. Other jackasses see that and imitate that behavior.I wonder if people doing this manage to create enough valid reports for it to be worth it (to them), or if it's just an endless stream of new people trying it with zero success.
The people who just ask blindly put code into an AI code scanner will also be the same people who ask the AI to create a proof of concept attack and then never test it.
Maybe a $5 submission fee could prevent these zero effort reports.
Honestly this seems like a good idea. The revenue raised should go into the pot used to pay out for genuine vulnerabilities.$5 might be too little considering how time consuming a bug report review can be. Make it $50. But then cURL team has to start doing taxes and whatnot
$5 might be too little considering how time consuming a bug report review can be. Make it $50. But then cURL team has to start doing taxes and whatnot
$5 might be too little considering how time consuming a bug report review can be. Make it $50. But then cURL team has to start doing taxes and whatnot
People who submit 20 false AI-generated vulnerability reports might also think that an AI can give good, free advice on a lawsuit to get back their $1,000.$5 might be too little considering how time consuming a bug report review can be. Make it $50. But then cURL team has to start doing taxes and whatnot
Windows Recall?I work for one of the large tech companies and can second the number of bogus reports we're getting is through the roof. The most egregious one I recall was an "attack vector proof of concept 9.8+" that was actually just a feature outlined in our public documentation.
In theory the companies hosting the bug bounty programs would be the collectors and distributors of payments so shouldn't be any extra admin on cURLs behalf.
Unless I'm severely misunderstanding something I'm not involved in bug bounties (not smart enough).
Edit: I really feel as though the onus falls on hackerone and similar to combat the flood. Otherwise someone will come along and take their lunch with a relatively simple change.