Open Source Taiwan: Securing the Open Source Supply Chain in the AI Era
Open Source Taiwan meetup with a serious lineup: a keynote from Brian Behlendorf (OpenSSF CTO, co-founder of the Apache HTTP Server), Justin McLean (ASF Incubator VP), and local panelists from EasyOpen Technology and MediaTek Innovation Hub. Topic: Securing the Open Source Supply Chain for the AI Era.
The size of the problem
Three numbers to set the scale: 96% of codebases contain open source components, ~74% of enterprise code is open source by composition, and npm gets more than 4,000 new packages every day.
The dependency chain is the vector: your app → framework → library → sub-library → a tiny utility maintained by a single person. That link can cascade vulnerabilities to thousands of downstream projects.
The main attack vectors: malicious packages impersonating legitimate ones, dependency confusion (exploiting resolution order between internal and public registries), typosquatting, and maintainer takeovers — the 2024 XZ Utils case was a two-year social engineering campaign to slip in an SSH backdoor. Alongside SolarWinds (2020, 18,000 organizations via the build pipeline), Log4Shell (2021), and PyTorch’s malicious torchtriton (2022), the pattern is clear.
And the phrase that ran through the whole event: “AI amplifies all of these — faster code, more attack surface.”
Project Glasswing
Behlendorf spent a good part of the keynote on Project Glasswing, the initiative Anthropic launched in April 2026: using frontier AI (Claude Mythos) to find and fix vulnerabilities in the world’s most critical open source software, before attackers do. It doesn’t just scan — it proposes fixes, so maintainers can act fast.
What struck me most wasn’t the technology but the equity angle: for the first time, a solo maintainer has access to the same caliber of security tooling as a large corporation. The coalition (AWS, Apple, Google, Microsoft, JPMorgan, IBM, Linux Foundation…) and the thousands of zero-days found back up the approach. As Jim Zemlin (Linux Foundation) put it, AI-augmented security should be “a trusted sidekick for every maintainer, not just for whoever can afford an expensive security team.”
The layered-defense model
The call to action, by level:
- Developers: pin dependency versions, run OpenSSF Scorecard on your projects, sign releases with Sigstore
- Organizations: mandatory SBOMs, SLSA levels in the build, contribute engineers to OpenSSF working groups
- Ecosystem: fund foundations and maintainers — “fund the foundation, not just the feature”
Note for my own stack: Spring Boot 3.x already makes it easy to generate SBOMs (CycloneDX) from Actuator. And personally, this meetup was part of the motivation for hardening this very blog’s dependencies: no lifecycle scripts on install, and a minimum 7-day package age — most malicious packages get caught and unpublished within hours.
The Ukraine → Taiwan analogy
The most geopolitical moment of the night: Ukraine discovered out of necessity that depending on foreign hardware (Chinese DJI drones) was a national security risk, and built its own industry with open source software + AI + local hardware. Taiwan faces the same equation with advantages: it makes the world’s best silicon, has deep technical talent — and an equally real geopolitical threat. Open source lets you audit the whole stack, reduce dependence on foreign vendors, and collaborate with allies.
The uncomfortable Q&A questions
The best Slido questions were left open, which is exactly why I’m noting them down:
- How do maintainers cope with AI slop, when burnout was already easy before AI?
- How do you measure open source ROI? (the eternal question for Taiwanese companies)
- “How do we know if there’s a person behind a contribution, or just another AI?”
That last question, I think, is going to define the next decade of open source.