Learn RAG prompt injection .NET defenses for hostile retrieved content: source admission, provenance, delimiters, least privilege, scanning limits, and tests.

I recently installed Ubuntu 26.04.1 on one of my laptops and wanted to add the GitHub CLI. I followed the official installation instructions and ran sudo apt install gh in the terminal.

But, of course there is a but; otherwise I wouldn’t be writing this post. The version installed was nowhere near the latest version from GitHub. I ran gh --version and got this:

gh version 2.46.0 (2026-06-18 Ubuntu 2.46.0-4ubuntu0.26.04.1~esm1)
https://github.com/cli/cli/releases/tag/v2.46.0

I went through the instructions again, purged the package, and reinstalled it. The same result.

Note how the output refers to 2.46.0-4ubuntu0.26.04.1~esm1. gh is not being installed from the GitHub repository, but from the Ubuntu repository.

To confirm this, run apt-cache policy gh and you will see something like this:

 1gh:
 2 Installed: (none)
 3 Candidate: 2.46.0-4ubuntu0.26.04.1~esm1
 4 Version table:
 5 2.101.0 500
 6 500 https://cli.github.com/packages stable/main amd64 Packages
 7 2.46.0-4ubuntu0.26.04.1~esm1 510
 8 510 https://esm.ubuntu.com/apps/ubuntu resolute-apps-security/main amd64 Packages
 9 2.46.0-4 500
10 500 http://us.archive.ubuntu.com/ubuntu resolute/universe amd64 Packages

Lines 5 and 6: Even though I have added the GitHub repository, it is at a lower priority (500) than the Ubuntu repository (510). Lines 7 and 8: The Ubuntu repository is the one that is being used to install gh.

I need to make the GitHub repository a higher priority. To do this, add a file called /etc/apt/preferences.d/github-cli with the following:

Package: gh
Pin: origin cli.github.com
Pin-Priority: 700

Now run apt-cache policy gh again, and you should see -

 1gh:
 2 Installed: (none)
 3 Candidate: 2.101.0
 4 Version table:
 5 2.101.0 700
 6 500 https://cli.github.com/packages stable/main amd64 Packages
 7 2.46.0-4ubuntu0.26.04.1~esm1 510
 8 510 https://esm.ubuntu.com/apps/ubuntu resolute-apps-security/main amd64 Packages
 9 2.46.0-4 500
10 500 http://us.archive.ubuntu.com/ubuntu resolute/universe amd64 Packages

Now, the GitHub repository has a higher priority, and when you try to install gh, it will be installed from the GitHub repository.

Run sudo apt install gh.

The Navier-Stokes problem starts with a cup of tea. Stir it, pull out the spoon, and watch the swirl slowly settle. The math that describes that swirl is about two hundred years old. For about ninety years, nobody could answer one question about it.

On September 8, 2026, OpenAI announced that its AI had found the answer.

The problem

The Navier-Stokes equations describe how fluids move. Water in a pipe, air over a wing, tea in a cup. Engineers use them to design airplanes and cars, and weather forecasts lean on them too. They even account for viscosity, the thickness that makes honey pour slower than water.

Picture a map covered in little arrows. Each arrow shows which way the fluid is moving at that spot, and how fast. The equations tell you how every arrow changes over time.

The famous question sounds almost too easy. If a flow starts out smooth, does it stay smooth forever? Or can the math break, with the speed at one spot shooting to infinity in a finite time?

That kind of break is called a singularity. In 2000, the Clay Mathematics Institute made the question one of its seven Millennium Prize Problems. A proof in either direction was worth a million dollars. Nobody collected.

The solution

OpenAI’s answer: the math can break. The company says about 10,000 AI agents worked on it for 88 hours.

Their example starts with fluid sitting perfectly still. A smooth outside force keeps pushing on it. A spinning core forms, then stretches and gets thinner. As it gets thinner, it spins faster, like a figure skater pulling in their arms.

A real skater runs out of arm. In the math, the core keeps shrinking, and within a finite time its speed has no limit. The total energy stays finite the whole time. All that speed piles into one shrinking spot. Viscosity keeps trying to smooth things out. It loses.

OpenAI also released a version of the proof in Lean, a language that lets a computer check every step. People still have to confirm that the statement it checked matches the real question.

Is it official?

Not yet. As I write this, the Clay Mathematics Institute still lists the problem as unsolved. Its rules say the proof must appear in a peer-reviewed journal. Then it has to hold up for two years before any prize.

OpenAI’s example also needs that outside force. The prize rules allow it, but many mathematicians say the version they care about most has no outside force. Expect that argument to run for a while.

And no, nothing in your kitchen is about to reach infinite speed. That only happens inside the equations.

One more thing before you go back to your tea. In my essay Verification Is the New Bottleneck, I argue that making the work is now the easy half. Deciding whether to trust it is the real job. The Navier-Stokes announcement is the same story, with a million dollars riding on it.

That essay is one of thirty in my book AI: Nobody’s in There. But we’re still in here. Every essay is free to read at pinaldave.com. There is also a paperback on Amazon, if you would rather hold something real.

The announcement is not the finish line, it is where the checking starts.

Published by Pinal Dave on SQLAuthority. More of my work at pinaldave.com.

First appeared on Navier-Stokes Problem: What It Is and What AI Found

2026-09-17

• Custom agents can opt into repository instruction files (AGENTS.md, copilot-instructions.md, CLAUDE.md) by setting include-custom-instructions: true in their frontmatter.
• Resuming an active session without plugin-directory, discovery, or working-directory overrides preserves marketplace plugins and skills after reload. Configuration read or validation failures no longer discard active plugins; missing-file and intentional-removal behavior is unchanged.
• /sandbox policy now reports local-network access using your configured setting
• The status row now says it is waiting for background shells, instead of "Working", when a turn ends while an attached background shell such as a dev server is still running.
• Resume sessions even when transcript files contain recoverable corruption
• Expanded reasoning text in the compact timeline is no longer dimmed, so it is as readable as the rest of the timeline.
• Autopilot stops after accepted task completion instead of continuing unexpectedly

My East Coast travel is nearly done as I jet to Texas to visit my kiddo in college. No reading list tomorrow as I navigate the Texas heat and BBQ.

[article] What happens to engineers when AI writes all the code? This piece calls out five places where engineers will spend time in the future. Now, for many.

[blog] Agent Anomaly Detection, now in Private Preview on the Gemini Enterprise Agent Platform. I like it. Your agent may be doing what it’s allowed to, but something doesn’t feel right. This new service adds no runtime latency but keeps an eye on agent execution.

[blog] Brownfield Agentic Engineering. Good topic. I’d imagine that every consulting company has a playbook like this right now. Tackling existing codebases with agents isn’t the same as start from scratch.

[article] When Persuasion Is (and Isn’t) Manipulation. It can be a fine line. I’ve studied the topic of persuasion for years, and there are ways to use this power for bad purposes.

[blog] How to De-Slop an AI-Generated Codebase. Tighter instructions, better rules, and more context can help you keep slop out of your code.

[article] Salesforce and Nvidia’s new reasoning model is everything the AI labs should fear. It’s open, domain-specific, and doesn’t need a lot of your context. Interesting. I doubt this is the ENTIRE future of models, but it’ll be a part.

[blog] Agents: The new, New Kingmakers. Indeed, the moment is upon us. Product and motions are now built around the agent experience and what they need.

[article] What decides which tasks product managers are willing to delegate to GenAI? This has some advice for teams struggling to integrate AI into their existing workflows.

[blog] Elevating Antigravity agent skills, Part 3: Parallel subagents. Good advice here, including some anti-patterns to look out for.

[blog] AI Changed How Spotify Builds. What We Learned (and Fixed) About Quality at Higher Velocity. Spotify has some pretty absurd usage numbers. Keeping systems online at their levels is no joke. This post has some stats and lessons learned (so far).

Want to get this update sent to you every day? Subscribe to my RSS feed or subscribe via email below:

How To Write With An LLM

Rule Number One: You may not use a single word an LLM suggests to you.

[...] I think that as a form of intellectual personal protective equipment you should adopt the rule that any specific turn of phrase an LLM suggests is off limits. Be strict about the rule!

I won't let LLMs write content for my blog, but I use them for fact-checking, spelling and grammar and as an occasional thesaurus (see my proofreading prompt).

The rule to never use a turn of phrase suggested by an LLM feels good to me. The text has that weird smell to it, and it's also a good principle to help stay disciplined.

Later in this piece Thomas shows a screenshot of his personal LLM copyediting tool (see also this Twitter thread), and provides a prompt to help kickstart building your own.

Tags: thomas-ptacek, writing, ai, generative-ai, llms