One Vendor Away

The Anthropic leak isn't the story. Your single-platform dependency is.

The company that sells you AI safety just had its third leak in twelve months.

Not a hack. A missing line in a config file.

On March 31, Anthropic accidentally exposed 513,000 lines of unobfuscated TypeScript through a source map file bundled in their public npm package. The kind of thing a build pipeline checklist catches. Except theirs didn’t.

The response made it worse. Their DMCA notice executed against roughly 8,100 repositories, including legitimate forks of Anthropic’s own publicly released Claude Code repository. Developers who had never touched leaked code woke up to their repos disabled.

They retracted most of it. Access was restored. That part doesn’t matter.

Here’s what does.

Every LLM vendor is one bad release day away from something like this. Anthropic just happened to go first. OpenAI, Google, whoever you’ve built on top of — they’re all software companies moving fast, and fast companies make operational mistakes. That’s not cynicism. That’s just how software at scale works.

The mistake operators are making isn’t using these platforms.

It’s building as if any single one of them is infrastructure.

If your workflows can’t survive one vendor having a bad week, you don’t have a system. You have a dependency.

The fix isn’t complicated. Know which of your critical operations run through a single platform. Have a fallback. Build the orchestration layer that lets you swap when you need to, not after you’re already down.

This story isn’t about Anthropic. It’s about what happens when you find out which of your assumptions were load-bearing.

~ Audra ✌️

---

P.S. This is exactly why I built Content Hub OS the way I did. No single platform owns the logic. The orchestration layer sits above the tools, which means when one goes sideways, the system doesn't. I built it on my own operations first to make sure it actually holds. If you're rethinking how your stack is structured, that's what CHOS is for.