The rapid proliferation of AI-assisted vulnerability research has unleashed a torrent of low-quality security reports that are drowning software maintainers in noise. Instead of helping teams patch real flaws, many submissions are duplicates, lack proof of concept, or represent theoretical attack scenarios that cannot be reproduced. This flood is straining open source projects and commercial bug bounty programs alike, with some industry leaders warning that the current system is becoming unsustainable.

Linus Torvalds Sounds the Alarm

Linus Torvalds, the creator and longtime maintainer of the Linux kernel, recently described the project's security mailing list as "almost entirely unmanageable" due to the sheer volume of duplicate reports. In a note accompanying the latest kernel release candidate, Torvalds explained that when multiple researchers use identical AI tools independently, they often discover the same vulnerabilities. "If you found a bug using AI tools, the chances are somebody else found it too," he wrote. He urged researchers to add genuine value by reading documentation, creating patches, and demonstrating real understanding rather than merely forwarding raw AI output.

Torvalds' frustration echoes a broader sentiment across the open source community. The Linux kernel, one of the largest and most critical software projects in the world, relies on a relatively small group of maintainers who must now spend hours each week sorting through automated reports. Duplicate submissions not only waste time but also raise the risk that a legitimate, critical vulnerability might be buried under piles of irrelevant noise.

GitHub Imposes New Validation Requirements

GitHub, one of the largest platforms for code hosting and security research, has acknowledged the problem firsthand. Jarom Brown, Senior Product Security Engineer at GitHub, noted that while lowering the barrier to entry for security research is generally positive, his team has been inundated with submissions that fail to demonstrate any meaningful security impact. These include reports without proof of concept, theoretical attack chains that collapse under scrutiny, and findings already covered by GitHub's published ineligible list.

In response, GitHub announced that all AI-assisted findings must now be validated by the submitter before being sent in. A complete submission must include a working proof of concept that demonstrates exploitation potential and concrete security impact. Reports that cover known ineligible categories will be closed as "Not Applicable," which may negatively affect the submitter's reputation score on platforms like HackerOne. Brown also asked researchers to keep reports concise, warning that bloated, AI-padded submissions slow down triage and waste everyone's time.

These measures reflect an industry-wide struggle. Many bug bounty platforms, including HackerOne and Bugcrowd, are deploying their own AI tools to filter spam, but the arms race between AI-generated noise and AI-powered triage is far from settled.

Researchers Lose Motivation

The collateral damage extends beyond program administrators. Experienced security researchers, who have spent years building credibility through high-quality, original findings, are finding their efforts undervalued. Shubham Shah, co-founder of Assetnote and a respected vulnerability researcher, pointed out that organizations now take far longer to review legitimate reports and act on real flaws. This delays the feedback loop that keeps top researchers engaged and incentivized.

"The joy of reporting vulnerabilities to bug bounties is quickly dissipating," Shah said. He warned that unless platforms can effectively differentiate between spam and legitimate research, he and others may retreat to private vulnerability research and invite-only bounty programs. For the industry, losing veteran researchers means a net decrease in security quality, as those individuals are often the ones who discover the most impactful vulnerabilities.

Open Source Projects Feel the Brunt

While large corporations like Microsoft and Google have dedicated security teams that can absorb some volume, open source projects are particularly vulnerable. They typically rely on volunteer maintainers who have limited time and no financial backing to handle an avalanche of reports. The cURL project, a widely used data transfer tool, experienced this firsthand. Its lead developer, Daniel Stenberg, originally decided to stop accepting HackerOne submissions and eliminated monetary rewards for security reports, hoping to remove the financial incentive for submitting AI-generated junk.

Initially, cURL moved to receiving reports via GitHub and email, but that approach proved less effective. A month later, the project returned to HackerOne, but kept the decision to discontinue bounties. The result was surprising: the number of reports increased, their quality improved, and the rate of confirmed vulnerabilities surpassed pre-2024 levels. Stenberg noted that the nature of submissions had changed completely — the slop problem disappeared. However, he cautioned that the higher influx of legitimate vulnerability reports still poses a challenge. "This avalanche is going to make maintainer overload even worse," he explained. Without adding more volunteer maintainers, projects may struggle to handle an expanding backlog of valid reports.

Industry Responses and Future Directions

HackerOne acknowledged the problem and advised customers to refine their program scope and submission guidelines to reduce noise. The platform also offers AI-assisted triage tools paired with human oversight. Michiel Prins, Co-founder and Senior Director of Product Management at HackerOne, emphasized the importance of preserving signal quality so that maintainers can stay focused on fixing real issues. "Our focus is helping programs manage that shift with workflows that filter noise early, surface credible reports, and keep vulnerability management sustainable," he said.

The Open Source Security Foundation (OpenSSF) has also stepped in. Its Vulnerability Disclosures Working Group is seeking community feedback to develop best practices, policy templates, and guidance for maintainers on how to spot and handle AI-assisted submissions. The goal is to create a framework that helps volunteer-run projects, which lack the resources of corporate security teams, to triage efficiently without ignoring genuine vulnerabilities.

The broader implications extend beyond individual projects. The industrial scale of AI-powered vulnerability discovery raises questions about the sustainability of the entire bug bounty model. If every automated tool can generate thousands of potential findings, the traditional approach of offering monetary rewards for unique discoveries may need to evolve. Some projects may follow cURL's lead and remove financial incentives for vulnerability reporting altogether, relying instead on community goodwill and intrinsic motivation.

At the same time, AI tools themselves continue to improve. Future systems might be trained to filter out low-quality submissions automatically or to generate well-formed patches alongside reports. However, as the current crisis shows, technology alone cannot solve the human problem of maintainer overload. Without a cultural shift in how researchers use AI — from drive-by reporting to thoughtful, validated contributions — the flood of junk reports will only worsen.

The Linux kernel's mailing list, once a manageable channel for security discussions, now serves as a cautionary tale. Torvalds' plea for researchers to add "real value on top of what the AI did" encapsulates the challenge: AI can augment human effort, but it cannot replace the judgment, context, and responsibility that comes with genuine security expertise.

Source: Help Net Security News