Curl Took July Off to Escape AI Slop. The Open-Source Supply Chain Is the Founder Risk No One Prices In.

On June 15, Daniel Stenberg — curl's creator and lead maintainer for over 27 years — announced that the project would not accept vulnerability reports through its HackerOne form from July 1 to August 3, 2026. Anyone with a paid support contract still gets full service; the public GitHub issue and pull-request trackers stay open. But the security intake queue is closed, and the 8.22.0 release slipped two weeks to September 2 as a direct consequence.

Stenberg was blunt about why. The project has been "under a huge pressure for the last four months or so," and he does "not expect this deluge to be over." The maintainers, he wrote, will "use this time of less pressure to take in some extra air and to enjoy the summer." His one-line summary of the problem doubles as the thesis: "If you found it with an LLM, anyone else could have found it, and probably already have."

The flood is AI-generated security reports. Stenberg has estimated that roughly one in five incoming submissions is AI slop — confident, well-formatted reports that cite functions or code paths that do not exist. And he was not alone in stepping back: he openly invited other maintainers to join the "summer of bliss," and several — including the maintainer of the widely embedded libexpat XML parser — said they would take the same break.

Open a modern SaaS and count the dependencies. Your framework, your HTTP client, your JSON parser, your crypto, your image library — most of it is maintained by a handful of people, often one person, usually for free. curl alone is estimated to run in more than 20 billion installations. The person keeping it patched just told you he is exhausted.

Here is the thing: a maintainer's breaking point is now your operational risk. When a real vulnerability lands during a "summer of bliss," the triage that protects your users is slower — or paused. When a core maintainer burns out and walks, the project that anchors your stack can stall or get handed to someone with less context. None of that shows up in your uptime dashboard until the day it does.

The detail most coverage skims is an economics one. Before LLMs, writing a security report that looked credible took real skill and time — a natural filter. That filter is gone. An LLM now produces a polished, plausible, completely wrong vulnerability report in seconds, for free. But the cost of a human maintainer reading it, reproducing it, and proving it wrong is exactly what it always was.

That asymmetry is the whole crisis. curl once ran a healthy confirmation rate on incoming reports; by 2025, coverage suggests the share of reports worth acting on had cratered as slop poured in. When 80% of what hits your queue is noise that costs you an hour each to dismiss, the queue itself becomes a denial-of-service attack — no exploit required. Closing the intake for a month is not laziness. It is a rational response to an unwinnable math problem.

And notice who is doing the flooding. A lot of these reports come from people chasing bug-bounty payouts or clout, pointing an agent at a famous repo and forwarding whatever it emits. That is the same behavior we flagged when the AI slop backlash hit content: when generation is free, the scarce, valuable thing is human judgment — and the cost of skipping it lands on someone else.

You cannot fix open-source funding by yourself. But you can stop being part of the problem and start pricing the risk you have been ignoring. Four moves, none of them expensive.

curl will be back on August 3, and the slop will still be there waiting. Expect more projects to adopt scheduled intake pauses, paid-only security channels, and stricter "proof of reproduction" gates before a report is even read. Some will lean on AI to triage AI — using models to pre-screen the reports that other models generated — which helps at the margin but does not fix the underlying funding gap.

The founders who come out of this ahead are the ones who treat open source like the critical infrastructure it is: funded on purpose, mapped for risk, and respected enough not to spam. Your stack is built on the goodwill of people you have never paid. A maintainer just reminded the whole industry that goodwill is not infinite. Use the reminder before it is your dependency going dark.