in reply to daniel:// stenberg://

1. curl://-tailing the AI onslaught?
2. Any time-related metrics to show how AI is making things more inefficient (e.g. rolling period totals/increase in PRs/bug reports and how many are assessed as completely worthless). Some key examples that show bad development/understanding that a reasonable dev should know is stupid, but still gets submitted anyway.
in reply to daniel:// stenberg://

I'm thinking something punning on bugs and plagues, trolls equipped with AI bludgeons. Flattening CURL: Timewasting trolls and the AI bug report barrage.
Curl in the Time Of Plague: Responding to AI-generated bug reports DDOSing a critical project.

I'm interested in the security implications and ideas for practical measures. I think the human factors are interesting - what are the demographics of people generating this nonsense - how many bad actors are we talking about, who are they, and what are their motivations?

in reply to daniel:// stenberg://

Regarding question 2:

Some AI content is easy to identify. How did you identify AI content in recent cases where it was less obvious?

Does time wasted handling fake "bug reports" correlate with AI generating more believable content?

For what portion of incoming reports are you unsure if AI or not? Does this influence how you handle these?

How much does this topic cost actual human reporters, e.g. by reports accidentally being closed as AI slop or time spent to prove being human?

in reply to daniel:// stenberg://

Title: AI (Argumentative/Abusive Interferants)' impact on cURL's security program.

Details: A stunning rebuke against everything that executives and salespeople claim to be true about AI in open source security.

Also, since "AI" is becoming a dirty word that left a poor taste in the mouth and now the conmen are shilling a new buzzword: "Agentic", I'd love a clear line in any charts showing when "Agentic" became a term to show how it doesn't magically make security reports better.

in reply to daniel:// stenberg://

no clever title for you, but some ideas on things to cover:

- key stuff like the asymmetry of effort involved with generated reports vs. responding to them

- your thoughts regarding the sustainability of bug bounties in the face of LLM submissions

- maybe a timeline + data/graph on how many nonsense reports you get

- impact on you personally (motivation, interpersonal with contributors, etc.)

- how it's changed your views on tech, FOSS, and the security community in general

in reply to daniel:// stenberg://

ad 2: I think it would be great to - from a Maintainer standpoint - talk a bit what bug reporting / a security program is for, what you expect from maintainers and what your assumptions are (for example: "When someone finds and reports an issue, I assume I can ask them questions about the problem to get more specific details"). From that I would like a bit of a reflection on whether LLM based processes do actually fulfil the basic requirements of that
in reply to daniel:// stenberg://

Androids dream of curl security vulnerabilities.

I’d like the talk to contain statistics and/or timelines that explain the scale of the problem, when it started, if it grows linearly or exponentially. What things you’ve tried to reduce the poor submissions and your thoughts on what works best. I’d like to see the most ridiculous example and the example that wasted most time.

in reply to daniel:// stenberg://

Abstract:

In these days of "vibe coding" and chatbots, users ask AIs for help with everything. Asked to find security problems in Open Source projects, AI bots tell users something that sounds right. Reporting these "findings" wastes everyone's time and causes much frustration and fatigue. Daniel shows how this looks, how it DDoS projects and how totally beyond crazy stupid this is. With examples and insights from the #curl project.

----

Good enough maybe?

#curl
in reply to daniel:// stenberg://

- Smack My AI Slop (bonus points if you walk on stage to The Prodigy)
- I got 99 AI problems (and slop is all of ‘em)
- It’s a hard slop life
- Yesterday, all AI troubles were so far away
- Once Upon a Time in Sloppywood
- The Pursuit of Sloppyness
- Don’t Slop Me Now
- How do they slop me? Let me count the ways.
- The Perks of Being a Maintainer
- Bagderman or (The Pernicious Deceitfulness of AI Slopporters)
in reply to daniel:// stenberg://

- Asymmetry in generating bullshit versus verifying submissions.
- Bug bounty programs encourage volume vs accuracy. Especially if you get paid for AI slop findings that aren’t valid but the vendor didn’t take the time to verify all 20,000 filings.
- If you don’t know the domain, you cannot tell the difference between AI slop bugs and the real deal bugs. But you do it anyway because your cousin told you it’s a shortcut to money, fame, clout. So, make it cost $ for unvetted submitter
This entry was edited (1 month ago)