Published Soursen pieces tagged ai, page 2, with the standfirst and the beat. Every fact is checked against the source.
Security
A fake captcha got one employee to paste a command into their own terminal, and that opened the door to 2 Senate departments. The attack is a warning about a social engineering trick that is spreading. The bigger lesson is what Berlin did after it noticed.
AI & GPU compute
MIT researchers had GPT-6 Astra edit 0.016% of a small model’s weights, with no training, and its math score jumped from 7.9% to 48.1%. The number is real and the method is strange. The limit they found in the same paper is the part worth understanding.
Linux & open source
The kernel just took in nearly 20,000 commits in one cycle, a record. The reason is the same force straining its maintainers, AI. The release is the clearest look yet at what that help gives and what it costs.
Security
In October 2026 a major security firm posted 6 Linux kernel vulnerabilities, 2 of them rated 9.0, straight to the public mailing list with no fixes attached. The maintainers were furious. The real story is that responsible disclosure may be breaking down for open source, and AI is the reason.
Security
Playwright bundles its own Chromium that lags Chrome stable, and it launches with the sandbox off by default. For a human running tests that is a small risk. For an AI coding agent browsing the open web on your laptop, it is a bigger one. The fix takes 2 settings.
Developer tools
On 31 October 2026 one of the people who helped build modern AI put up a YouTube clone sourced from 20 years of Hacker News comments. He wrote almost none of the code himself. The build is a small thing. What it says about who can ship software now is not.
AI & GPU compute
Audited 2025 figures show a company tripling revenue while burning cash faster than any startup ever has. The loss is only half the story. The other half is a loop of deals that turns the same dollars into growth at three companies at once, and a bill that has four ways of reaching you.
Security
CVE-2026–98432 is a use after free flaw in the Linux kernel’s TCP code, rated 9.8 out of 10, that can hand a remote attacker root on a server running multipath TCP. What makes it different from the usual critical CVE is who found it. An unreleased internal AI model at Anthropic spotted the bug, then built a working root exploit in about 53 minutes. I went through what the flaw is, who is patched, and why the AI part changes the math for defenders.
Infrastructure
China ran a trade surplus of $938 billion in the first 9 months of 2026, installed 59% of the world’s new factory robots, and refines around 90% of the planet’s rare earths. The interesting question is not whether it is ahead in manufacturing, the numbers settle that, but how a few deliberate choices compounded into a lead that tariffs and chip bans have not reversed. I went through the mechanics.
Linux & open source
Intel and HP spent roughly $10 billion building Itanium, the chip that was supposed to replace x86 forever. Analysts projected $38 billion in sales by 2001. In Linux 6.7, released in January 2024, the kernel deleted about 65,000 lines of Itanium support in one go, because nobody was left to maintain it. I went through why the most expensive bet in CPU history ended as a pile of deletions, and what it taught Linus Torvalds about anyone who promises the future needs a clean break.