The Infrastructure is Cracking—and Nobody's Ready

---

Ubuntu down for a day. A supply-chain attack that somehow knew exactly which security firms to target. A million-download package stealing credentials like it’s 2015. Healthcare sites leaking citizenship data to ad tech companies. This isn’t a crisis yet. It’s the warning light you ignore until the engine catches fire.

The past two weeks have exposed something I’ve been half-expecting for years: we’ve built the entire internet on foundations that nobody actually maintains. And the people trying to exploit those gaps are getting smarter about it.

Photo by Maarten van den Heuvel / Pexels

When the Basics Break Down

Let me be blunt about Ubuntu going down for over a day. That’s not supposed to happen. Ubuntu powers an enormous slice of cloud infrastructure—AWS instances, Docker containers, build systems for basically every major tech company. One day doesn’t sound catastrophic until you remember that “one day” in 2017 meant AWS S3 taking down a chunk of the internet and costing companies millions. The difference is S3 was transparent about it. We still don’t have a clear picture of what happened with Ubuntu’s infrastructure beyond “it was down.”

That opacity is worse than the outage itself.

Then there’s the Linux vulnerability everyone’s calling the most severe in years. I’m not going to pretend I can detail the exact technical specifics without seeing the actual CVE details, but the pattern is crystal clear: critical infrastructure is getting hit by vulnerabilities that should’ve been caught in code review. The fact that security firms like Checkmarx and Bitwarden were deliberately targeted in recent supply-chain attacks suggests someone’s doing reconnaissance. They’re not just throwing exploits at the wall—they’re studying which companies would maximize their access if compromised.

That’s a shift from spray-and-pray attacks.

An open-source package with 1 million monthly downloads—let that number sit for a second. One million. That’s not a niche library. That’s something running in production at banks, healthcare companies, financial services. And it stole credentials. We know this happens, but the scale keeps getting worse because the incentive structure is broken. Maintainers get paid zero dollars to keep their packages secure. Users get zero visibility into whether the packages they depend on are actually maintained. Someone comes along, buys the package from a burned-out maintainer, or just takes over an abandoned repo, and suddenly you’ve got malware in the supply chain.

This is the digital equivalent of selling your house and not telling the new owners that the foundation has termites.

Photo by UMA media / Pexels

The Enterprise Land Grab Is Ugly

Meanwhile, Sierra just raised $950 million—now sitting on over $1 billion—to become the “global standard” for AI-powered customer experiences. Anthropic and OpenAI are both launching joint ventures with asset managers to sell enterprise AI more aggressively.

Here’s what’s happening: the companies that won the AI wars (OpenAI, Anthropic, maybe Mistral) are now pivoting from research and chatbots to extracting money from every large organization on Earth. That’s not criticism—that’s how venture capital works. But it means they’re going to be ruthless about moats. They need proprietary datasets, custom models, lock-in architecture. They can’t let their enterprise customers actually own their data or models.

My read: in 12 months, you’re going to see enterprise AI adoption that’s real but also creates the exact conditions for the next major breach. These companies are integrating AI into customer service systems, internal tools, decision-making software. They’re doing it fast because the competition is doing it fast. Security is getting bolted on after the fact, if at all.

Sierra’s pitch is about “control” for enterprises. But control over what? If your customer experience engine is running on infrastructure you don’t own, trained on data you’re not sure about, with models you can’t audit, you don’t have control. You have a subscription bill and a vague hope that Sierra isn’t cutting corners on security.

The Open Source vs. Closed Source Knife Fight

Elon texted OpenAI’s president saying Sam Altman and Greg Brockman “will be the most hated men in America.” This is theater, obviously. But it’s theater that reveals something real: the open-source AI crowd (Elon, xAI) thinks the centralized approach (OpenAI, Anthropic) is dangerous. They’re not wrong. They’re also not trustworthy. But they’re not wrong.

Acorn just launched as a decentralized alternative to Twitter’s Communities, giving organizations control over their own spaces. That’s technically interesting and strategically pointless until someone figures out how to monetize it. But it’s a signal: there’s real appetite for infrastructure that doesn’t depend on a megacorp deciding whether you’re allowed to have nice things.

The irony is that decentralized systems are harder to secure. You can’t centrally patch a vulnerability if every node is independent. But they’re also harder to exploit at scale because there’s no single point of failure.

Here’s what I think is actually happening: we’re slowly realizing that “choose one: centralized and fast, or decentralized and slow” was always a false choice. What we actually need is federated—centralized enough to be manageable, distributed enough to be resilient. Ubuntu could have been. It wasn’t.

Photo by Denys Gromov / Pexels

The Data Leakage That Nobody Will Face Consequences For

Virginia and Washington, D.C. health insurance marketplaces were sharing citizenship and race data with ad tech companies until Bloomberg investigated and someone finally hit the brakes. Not because it was illegal (it basically wasn’t), but because it was embarrassing.

This one actually makes my blood warm in a bad way. Healthcare marketplaces are state-run systems that people access because they legally have to. They’re not choosing to share their race with advertisers. And somehow, somewhere, someone at a state health department decided that relationship was worth money. Or didn’t think about it at all. Which might be worse.

I’d bet we find out in six months that this was happening in at least five other states. And I’d bet that even after it’s discovered, the fines will be small enough that the cost-benefit analysis for future data sharing still works out in favor of sharing.

This is why I don’t trust enterprise AI to protect sensitive data. The incentives are all wrong. The people running these systems don’t get fired for leaks. The companies that buy the data don’t get punished. Only the users get harmed, and they didn’t consent to any of this.

What I’m Watching

• Ubuntu and Linux maintenance models, specifically how Canonical responds to the infrastructure downtime. If they treat it as a one-off incident rather than a signal that their maintenance infrastructure is under-resourced, we’ll see this again. Watch for public post-mortems and resource commitments by end of Q2 2024.

• Enterprise AI contract terms. When Anthropic and OpenAI start signing major customers, pay attention to whether those contracts include data residency guarantees, audit rights, and explicit security SLAs. If they don’t, the vulnerabilities in those enterprise AI systems will be catastrophic when (not if) they’re exploited.

• Open-source package takeover patterns. Security researchers are starting to track when maintainers transfer repos or packages to new owners. If you see a spike in transfers of widely-used packages, that’s a leading indicator of incoming supply-chain attacks. The next big one probably happens in the next six months.

• State-level healthcare data sharing scandals. Expect at least three more states to get caught doing this by mid-2024. The interesting question is whether any of them actually change their practices or just get better at hiding it.

The infrastructure isn’t cracking yet. We’re just finally noticing all the cracks that were already there.