AI Software Security: A Month of Breaches in One Day

8 min read

Came here from Instagram, LinkedIn, X or YouTube? Here is the full source list you came for. Everything below is built from these, nothing else.

The newsletter

• The Pulse: Time to rethink software security because of AI?, Gergely Orosz, The Pragmatic Engineer, Oct 22 2026

On October 22, 2026, Gergely Orosz opened his weekly issue of The Pulse with a blunt question: is it time to rethink software security because of AI? His reason was a stretch of days he called "absolutely wild", with security breaches and major vulnerabilities exposed across tech worldwide. Many of them landed on the Tuesday and Wednesday of that week, October 20 and 21.

His framing is the part worth remembering: a month's worth of security breaches in a single day, many of them fuelled by AI. Attackers are gaining an advantage by using AI, and defenders will have to respond. The same issue also covered an accusation that Anthropic's new version control system is a "rip off" of an open source project, plus a set of smaller stories, including a gzip speed boost for Claude Code.

Key takeaways

• The Pulse described a month's worth of security breaches in a single day, with many incidents on Tuesday October 20 and Wednesday October 21, 2026.
• Orosz's read: many were fuelled by AI, attackers have the advantage right now, and defenders have to respond.
• The issue looks closely at 3 of the incidents. That deep dive is for paid subscribers, so I am not naming them here.
• The creator of an open source, high-throughput version control server accused Anthropic of copying their design and presenting its product as new without credit.
• Also in the issue: Claude Code got a major speed boost with gzip, Shopify is building tooling for self-healing software, and Amp cofounder Thorsten Ball predicted that open source no longer makes sense.

What actually happened this week?

Here is what the source says, and only what it says.

Story	What The Pulse reported
Security	A series of high-profile incidents and breaches in a single day, many fuelled by AI. Many took place on Tuesday and Wednesday.
Anthropic	The creator of an open source, high-throughput version control server says Anthropic ripped off their design and did not credit the existing project. Orosz: "It's not a good look for the leading AI lab."
Saudi Arabia and OpenAI	Saudi Arabia made a "problematic" offer to pay the bail of an alleged OpenAI spy.
Shopify	Building tooling for self-healing software.
Foursquare	Its sale disappointed.
Claude Code	A major speed boost with gzip.
Predictions	Thorsten Ball, software engineer and Amp cofounder, made bold calls, including that open source no longer makes sense.

One honest note before we go further. The free part of the issue ends right after the security summary starts. The names of the three incidents, and the details of each, sit behind the paywall. I have not seen them, so I will not guess. If you saw a post claiming to know exactly which companies were hit, check whether it links to anything.

Why does this matter to a developer?

Because the problem Orosz describes is about cost, and cost decides how often you get attacked.

Think about what an attacker used to need. Time to read your code, time to write a convincing phishing email, time to try one more variation of an exploit. Every one of those steps had a human hour attached to it. If AI is doing a share of that work, as Orosz says it is in many of these incidents, the hour shrinks. When trying gets cheaper, people try more.

That is how you get a month's worth of breaches in one day. Not a new class of genius attacker. Just more attempts, landing at once.

Now look at it from your side. Most of us do not work on a security team. You are the developer who owns a service, a side project, a repo with a few secrets in CI, maybe an API key for an AI tool sitting in a .env file. You were never the target of a dedicated human attacker because you were not worth the hours.

If the hours stop mattering, that protection goes away. The small target becomes a target simply because it is there.

The second half of his sentence is the one I care about: defenders will have to respond. He does not say defenders are doomed. He says the response is not optional anymore.

Is the Anthropic version control story a security story too?

Not directly, but it belongs in the same conversation.

The claim is simple. Someone built an open source, high-throughput version control server. Anthropic shipped a version control system. The original creator says Anthropic took their design and presented its product as novel, without crediting the project that came first.

I am not going to rule on who is right. The free part of the issue does not include Anthropic's side, and I only write from what I can read.

What I will say is this. Open source runs on credit. Credit is how maintainers get jobs, sponsors and the energy to keep going. When one of the biggest AI labs is seen as taking without crediting, maintainers have less reason to publish in the open.

That connects to the last item in the issue. Thorsten Ball's prediction that open source no longer makes sense is a bold call, and Orosz labels it as one. Put it next to the Anthropic accusation and you can see why someone might make it.

And here is the security angle. Open source code is the code you can audit. A world with less of it is a world where you trust more black boxes.

What changes in practice?

Nothing in your stack changed on October 20. What changed is the assumption you should work from.

The old assumption: "Nobody is going to bother with my little service."

The new assumption: "Something will try my little service, often, and it will not get tired."

That shift changes a few habits.

1. Secrets age faster. A token that sat in an old commit for two years was a quiet risk. If scanning and trying are cheap, quiet risks get found.
2. Your AI tools are part of your attack surface. If you give an agent access to your repo, your terminal or your cloud account, that access is now something worth stealing. Treat agent permissions the way you treat a new teammate's: only what the job needs.
3. Patching speed matters more than it did. If attackers move faster, the gap between "a vulnerability is public" and "someone tried it on you" shrinks. Slow update cycles cost more.
4. Defenders get the same tools. Orosz's point cuts both ways. If AI helps attackers find holes, you can point it at your own code and find them first.

Notice the Shopify item in the same issue: tooling for self-healing software. The free text gives no details, so I will not invent any. But the direction is clear enough. The answer to faster attacks is faster repair.

What should you do this week?

Five things. None of them need a security team.

• Rotate one long-lived secret. Pick the oldest API key or personal access token you own and replace it. Then revoke the old one, not just stop using it.
• List what your AI agents can touch. Write down every tool you have given repo, shell or cloud access to. Remove anything you do not use weekly.
• Turn on secret scanning for your repos if your Git host offers it, and look at what it finds in your history, not just new commits.
• Update dependencies in the one project you have been putting off. Read the changelog for anything marked as a security fix.
• Ask your AI assistant to attack your own code. Give it one endpoint and ask how it would break it. Fix the first real thing it finds.

If you are a paid subscriber to The Pragmatic Engineer, read the full issue for the three incidents Orosz covers in depth. That is where the specifics are, and specifics are what you want before changing anything bigger than the list above.

FAQ

What did The Pulse say about AI and software security?

Gergely Orosz wrote that a series of high-profile security incidents and breaches happened in a single day, many fuelled by AI. He described it as a month's worth of breaches in a day, with many on October 20 and 21, 2026. His conclusion is that attackers are gaining an advantage by using AI and defenders will have to respond.

Which companies were breached?

The issue looks closely at three incidents, but that section is for paid subscribers. I have not read it, so I am not naming any companies. If you see a list of names, check that it cites a source.

What is the Anthropic version control "rip off" accusation?

The creator of an open source, high-throughput version control server accused Anthropic of ripping off their design and presenting its new version control product as novel without crediting the existing project. Orosz called it "not a good look for the leading AI lab." The free part of the issue does not include Anthropic's response.

What else was in this issue of The Pulse?

Saudi Arabia made a "problematic" offer to pay an alleged OpenAI spy's bail, Shopify is building tooling for self-healing software, Foursquare's sale disappointed, and Claude Code got a major speed boost with gzip. It also covered predictions from Amp cofounder Thorsten Ball, including that open source no longer makes sense.

Rotate one secret today, then check the free tools on Art of Code for the boring parts of your workflow. For more on where AI is taking developer work, browse the AI section of the blog.

Written by Adithya Guttha, Founder of Art of Code.