cURL is shutting down its vulnerability reward program after maintainers were hit by a wave of poor security reports, including reports generated with large language models that described vulnerabilities which did not exist.
The decision affects one of the Internet’s most widely used networking tools. cURL is used by admins, researchers, security professionals, and others for file transfers, troubleshooting web software, and automation.
Why cURL is ending the reward program
Daniel Stenberg, the founder and lead developer of the open source app cURL, said the project is too small to absorb the growing burden created by bad reports.
“We are just a small single open source project with a small number of active maintainers,”
He added that the team cannot change the behavior of people using AI systems to produce weak submissions. His explanation was direct: “It is not in our power to change how all these people and their slop machines work. We need to make moves to ensure our survival and intact mental health.”
The termination was made official in an update to cURL’s official GitHub account. According to the source article, it takes effect at the end of this month.
The move is not only about inconvenience. For a small open source project, every report has to be read, assessed, and answered. A false vulnerability claim can still consume maintainer time, especially when the submitter insists the problem is real.
What the bad reports look like
The problematic submissions described in the source are not simply rough drafts or incomplete findings. Some involved vulnerabilities that appeared to have been invented by an AI system.
Stenberg has posted a page listing specious reports submitted in recent months. In one response to such a report, a cURL project member wrote: “I think you’re a victim of LLM hallucination.”
The same response pointed to several warning signs. The report resembled other bogus material, included a code snippet of “curl_easy_setopt” that did not match the real function signature, contained code that would not compile, and referred to a changelog that did not match reality.
After the reporter complained and repeated the claim, Stenberg responded: “You were fooled by an AI into believing that. In what way did we not meet our end of the deal?
In a separate post on Thursday, Stenberg also warned that the project would take a hard line against time-wasting reports: “We will ban you and ridicule you in public if you waste our time on crap reports.”
Why this matters for cURL security
cURL’s importance makes the decision notable. The tool was first released three decades ago under the name httpget and later urlget. It is now integrated into default versions of Windows, macOS, and most distributions of Linux.
Because cURL is used to interact with large amounts of data online, its security process matters. Like many software projects, cURL has relied on private reports from outside researchers. The reward program gave cash bounties for reports of high-severity vulnerabilities.
Some cURL users objected to ending the program because they saw it as treating the symptoms of AI slop rather than the cause. They were also concerned that removing the bounty system could weaken a useful channel for finding and maintaining the security of the tool.
Stenberg largely agreed with the concern, but said the team had little choice. The issue, as presented by the project, is capacity: useful vulnerability disclosure depends on maintainers being able to separate real problems from noise without being overwhelmed.
AI-assisted research is not the same as AI slop
Stenberg has not rejected AI-assisted security work across the board. In September, he publicly praised researcher Joshua Rogers for sending a “massive list” of bugs found using AI-assisted tools. At the time, that report had resulted in 22 bug fixes.
In an interview, Stenberg said Rogers mostly used the AI-powered code analyzer ZeroPath. He described that kind of work as “A clever person using a powerful tool,” drawing a line between responsible tool use and careless AI output.
His criticism is aimed at a different pattern: people asking an AI bot for a vulnerability report, then sending the result without understanding whether it is true. Stenberg wrote: “I believe most of the worst reports we get are from people just asking an AI bot without caring or understanding much about what it reports.”
That distinction is central to the cURL decision. AI can help a capable researcher investigate code, but AI-generated text can also create a persuasive-looking false report. For a maintainer, both arrive as work that must be checked.
A warning for bug bounty programs
The cURL case may point to a wider problem for software security. The source compares the situation with music-streaming services, where AI slop has flooded platforms with songs, sometimes misattributed to real artists, making discovery harder.
Bug bounty programs depend on signal. If the number of low-quality vulnerability reports rises sharply, the cost of review can increase even when the number of real bugs does not.
For cURL, the answer is to end the reward program and protect the small maintainer team from a workload it can no longer justify. The security challenge remains, but the project is choosing a process it believes it can survive.