Rethinking how we do open source

Why this conversation

When models find vulnerabilities faster than people can patch them, the question is where security starts. Kalle Westerling is a Staff Developer Relations Engineer at Chainguard. The host recorded with him on 30 October 2026 at Dutch Cloud Native Day, at the Jaarbeurs in Utrecht.

A new security flaw every day

“Every day I wake up with a new security flaw”, he said. Frontier models are discovering more and more vulnerabilities, “both in new releases, in old releases, in end-of-life releases”, in the fabric of critical infrastructure built on top of open source: “things are moving at a pace and at a depth that we haven’t seen before.” It creates, in his words, “that sort of feeling in the morning of waking up and saying, oh no, another one that we haven’t seen ever.” He named the big Trivy break, “only a few months ago”, and the big worms, Shai-Hulud and Shai-Hulud 2.0. The attackers have access to the same kind of models, so there is “this acceleration driven by AI in discovering zero days”. Part of the defence is “trying to stay as vigilant as possible”: do your research, read up, follow the good sources.

Start left

At Chainguard the idea is that it is too late to shift left: “You have to start left.” Rather than mitigating things along the way, “we need to really rethink how we do security from the get-go.” The company has “a suite of tools and build essentially all of open source from source”, with its own harness to fix and patch zero-day vulnerabilities that do not have an exposure in the wild yet: “we’re trying to make sure that we’re mitigating these things at the very source code that we work with.” Customers get container images and library dependencies in Python, Go, JavaScript, Java and .NET, and more ecosystems are coming. “So we’re essentially rolling our own OS”, which brings rolling updates: no waiting for the big releases of bigger distributions.

CI workflows and agent skills

The CI/CD workflows are another attack surface: the dependencies they pull in get access to your keys, and from there to production and staging systems, so Chainguard provides hardened versions for them too. Then there are agent skills. Markdown files themselves can expose you to vulnerabilities, and skills can include scripts: “you really need to rethink the way that you’re doing AI development altogether”, from library dependencies and container images in production to the agent workflows and the CI/CD workflows that run them.

Getting started

Getting started is quite easy, he said. On Chainguard’s website you can sign up for five free images from the paid catalog: “You just fill out a form on the site.” And a free catalog of 60 images needs no authorization at all. Self-serve access to the libraries is rolling out soon. “We have over 3,000 images right now”, he said, with libraries backported in Python, Java and JavaScript and CVE remediation rolling out. They are hiring a lot too. “I love my job.”

When it comes out

The episode is edited and in the release order. Like every episode it goes out on YouTube first, then on Apple Podcasts, Spotify and Amazon Music.