The Open Source Security Reset

Finding bugs is becoming cheap. Proving fixes reached production is the scarce part.

That is why reset is the right word.

For years, open source security was treated as a visibility and funding problem: more money, more scanners, more training, better defaults. That work still matters. It did not create enough repair capacity.

GitHub's Secure Open Source Fund is the cleanest example. It moved real money and education toward maintainers. It also showed the limit of the old playbook: help projects improve, then hope the repair queue becomes manageable.

AI discovery at scale creates a remediation queue, not a curriculum. Valid findings still need proof, review, disclosure timing, upstream patches, downstream adoption, and evidence that production systems moved. That is not a training problem. It is an engineering capacity problem.

Akrites moves the work into coordination

Akrites starts from the uncomfortable premise that AI has changed vulnerability discovery faster than maintainer capacity. Its member list matters because it gives the project reach across infrastructure, cloud, finance, security, and open source foundations. The operating model matters more: intake, remediation, and disclosure coordination sit in one place.

The goal is not more public vulnerability announcements. The goal is controlled repair before a bug becomes a race between exploit builders and maintainers. Once a patch is public, the diff can become exploit guidance. If AI accelerates reverse engineering, the time between advisory and exploitation gets shorter. Coordination becomes part of the control.

Patch the Planet is the repair side

Patch the Planet starts where reports stop. OpenAI backs the initiative, and Trail of Bits supplies security engineers who work with critical open source projects to find bugs, write patches, and move fixes upstream.

The first week made the shape concrete: hundreds of discovered bugs, 64 pull requests, 51 issues across 19 projects, and 19 issues already closed with fixes. The listed projects are not peripheral. They include cURL, NATS, pyca, Sigstore, aiohttp, Go, freenginx, Python, PyPI, urllib3, Valkey, and RustCrypto.

A vulnerability report only starts the work. The hard part is proving the bug, choosing the fix, testing the patch, avoiding breakage, and helping the maintainer ship safely. AI makes discovery scale. It does not remove ownership.

Security companies can fail in the layer they sell

Trivy and KICS matter because they move the argument beyond underfunded maintainers. Trivy is tied to Aqua Security. KICS is tied to Checkmarx. These are companies whose credibility comes from code security, scanning, cloud security, and supply chain controls.

Aqua's Trivy incident account said the project and its users were hit by a cyberattack and that Aqua failed to protect its supply chain to the standard users expected. The timeline included stolen credentials, malicious artifacts, Docker Hub releases, credential resets, repository hardening, and outside review. Endor Labs connected related activity to TeamPCP and placed Trivy and Checkmarx's KICS in the same nearby security target set.

That is the cynical part. The failures happened in the layer these companies build products around: credentials, repositories, CI/CD, extensions, container images, distributed artifacts, and release governance.

This is not a branding problem. A scanner or IaC analyzer often sits inside the build and release path, where a compromise can touch source code, credentials, artifacts, and automation. Understanding flaws in other people's code is not the same as controlling the path that turns your own source code into trusted artifacts.

Security tools do not get a lower bar because part of the product is open source.

Repair capacity is the gap

SBOMs, CVEs, reachability, SCA alerts, policy gates, and upgrade recommendations still matter. They tell teams what exists and what is broken. They do not guarantee that an upstream project can validate, patch, release, and prove adoption fast enough.

Akrites and Patch the Planet point at the gap: upstream response capacity.

Someone has to validate the finding, write the fix, coordinate disclosure, identify affected products, produce mitigations, and verify deployment. Automation can help pieces of that work. It does not remove ownership.

The reset is not about replacing maintainers with AI. It is about admitting that discovery has scaled faster than repair. GitHub's fund improved the old model. Trivy and KICS showed even security companies can fail in their own release path. Akrites and Patch the Planet move the answer toward the work that decides whether a fix reaches users.

The measure is no longer alert volume. It is repair capacity.

The test is whether reviewed fixes actually reach production.