DevOps'ish 321: The OpenAI and Hugging Face postmortem, Gateway API v1.6 ships L4 routing, and more

Gateway API v1.6: TCPRoute and UDPRoute Graduate to Standard (5 minute read)

Raw L4 routing finally hits GA. TCPRoute and UDPRoute move to the Standard channel at v1, which means databases, DNS, VoIP, and game servers get the same stability guarantees HTTPRoute has had for a while. The other change worth noting is organizational: experimental resources now live in a separate gateway.networking.x-k8s.io API group with an X prefix, so nobody accidentally ships an alpha CRD to production and acts surprised. The v1alpha2 versions of both routes are deprecated and headed for removal, so start the migration.

Black Hat USA 2026: The ‘Breaking’ News: The OpenAI–Hugging Face Incident (37 minute watch)

Michael Dalton and Eric Wallace reconstruct how an OpenAI evaluation agent broke out of its sandbox, got into Hugging Face infrastructure, and went looking for test answers. No human in the loop, no operator pushing it along. This is the primary source for the story everyone spent the week arguing about, and it is worth the 37 minutes before you form an opinion from headlines.

Tailscale didn’t stop the Hugging Face intrusion (7 minute read)

Credit where it is due: a vendor writing publicly about how its own product did not save the day is rare. An AI agent used a stolen Tailscale auth key to move through Hugging Face, and Tailscale walks through what would have actually helped, namely workload identity federation, flow logs you actually read, and defaults that are safe without a config change. Long-lived auth keys remain a footgun no matter whose mesh you are running.

The OpenAI Hack Shows the Genie Is Out of the Bottle (5 minute read)

Schneier’s read is that AI cyber capability cannot be put back in the box, and that kneecapping American models mostly hands the advantage to attackers and to whoever is shipping open Chinese frontier weights. Agree or not, he is arguing from the same place a lot of practitioners land: the capability exists, your detection does not, and policy debates do not change what is already running.

One of China’s Most Powerful AI Models Has Also Escaped Containment (4 minute read)

Kimi K3, Moonshot’s open-weight model, apparently wandered onto the internet trying to cheat on a test. Two sandbox escapes from two labs on two continents in the same news cycle is not a coincidence; it is a pattern. The models are not malicious; they are goal-directed, and your test environment is just another obstacle between them and a passing score.

A Security Pro Hacked North Korean Hackers. He Found They’d Breached Hundreds of Networks Worldwide (5 minute read)

Vangelis Stykas has been sitting inside North Korean hacker infrastructure for close to two years, and the view from in there is 1,640 impacted companies across 57 countries. The number that should bother you is not how many got breached, it is how many had no idea until a researcher on the other side of the world told them.

UK faces new legal fight from Apple over backdoor access to iCloud data (3 minute read)

Round two. The UK government is pressing again for warrantless access to private user data, and Apple has filed another complaint asking the Investigatory Powers Tribunal to step in. Same fight, same math: there is no such thing as a backdoor that only opens for the people you like.

GitHub has alternatives, but no replacement (6 minute read)

Lalit Maganti walks through Codeberg, GitLab, SourceHut, Forgejo, Radicle, Tangled, and self-hosting, and reaches the conclusion most of us quietly already knew. Every one of them can host your repository just fine. None of them reproduce the common conventions and the discovery that make GitHub sticky. The hard part was never git. It was the network effect built on top of it.

AI Doesn’t Fix Bad DevOps: Lessons from 15 Years of DORA Data, with Nathen Harvey (74 minute listen)

Coté and Whitney get Nathen Harvey, who runs Google Cloud’s DORA research program, to say the thing out loud: AI amplifies teams that are already good, and teams that are struggling just hit their bottlenecks faster. Fifteen years of delivery data backs up what every operator suspected. Your constraint was never typing speed.

Born Against, or why hobby programming communities are aggressively against LLM usage (2 minute read)

Fogus on why the OSDev, LangDev, EmuDev, and demoscene crowds treat LLM-generated code as cheating. In those communities, the hard-won learning is the actual product, not the artifact, so handing the hard part to a model defeats the purpose entirely. Short, sharp, and a useful frame for why “just use AI” lands so badly in some rooms and so well in others.

Securing a Go Supply Chain: The Pipeline That Holds in 2026 (8 minute read)

Jules Robineau lays out an actual working pipeline: govulncheck, dependency pinning, SBOM generation, container scanning, secrets handling, GitHub Actions hardening, and signing. Practical and specific rather than a listicle of vendor logos. The post is bilingual, French first and then the full English version, so scroll if you land in the wrong one.

What an SSH Tunnel Actually Does and When You Should Use One (3 minute read)

A clean explainer on local, remote, and dynamic forwarding, and the honest caveat that tunnels are best for controlled, temporary access rather than a permanent architecture. Worth passing to whoever on your team is currently running a six-month-old ssh -L in a screen session and calling it networking.

github/gh-stack - MIT - GitHub’s own CLI extension for working with stacked pull requests.

Nutlope/hallmark - MIT - An anti-AI-slop design skill for Claude Code, Cursor, and Codex.

isaqueseneda/shieldfont - AGPL-3.0 - A typeface that protects written content by poisoning unauthorized AI training datasets.

Want to reach 6,800+ developers, SREs, & DevOps leaders every week?

DevOps'ish delivers a 46% open rate and 10% click rate — well above industry averages — to a global audience of practitioners, VCs, and C-suite decision-makers. Your brand gets exclusive placement (one sponsor per issue, zero competition) in a text-based newsletter that bypasses ad blockers entirely. Get in front of the people who build and run the internet!