Ghostty's Terminal Engine Hits 1.0, and Neovim Is Already Moving In

by Sevda Polat — November 7, 2026

Mitchell Hashimoto shipped libghostty-vt 1.0 on Friday. It's the engine inside the Ghostty terminal, now available as a portable, MIT-licensed C library with its own release schedule and slimmer tarballs. The headline feature is a stability promise: semantic versioning and strict ABI compatibility, so apps that embed the library can upgrade without breaking. Neovim has already approved a pull request to replace libvterm, its current terminal library, with it, which is about as strong a vote of confidence as a terminal library can get. A 1.0 like this shifts the cost of breakage from every downstream app onto one maintainer, so who owns Ghostty matters more than the version number.

Ghostty's Terminal Engine Hits 1.0, and Neovim Is Already Moving In

by Sevda Polat — November 7, 2026

Mitchell Hashimoto shipped libghostty-vt 1.0 on Friday. It's the engine inside the Ghostty terminal, now available as a portable, MIT-licensed C library with its own release schedule and slimmer tarballs. The headline feature is a stability promise: semantic versioning and strict ABI compatibility, so apps that embed the library can upgrade without breaking. Neovim has already approved a pull request to replace libvterm, its current terminal library, with it, which is about as strong a vote of confidence as a terminal library can get. A 1.0 like this shifts the cost of breakage from every downstream app onto one maintainer, so who owns Ghostty matters more than the version number.

AI security week

Happy Saturday. The security headlines today sound new, but most of the ideas underneath them are older than the people shipping the products. A little history helps:

1988 named the "confused deputy." Computer scientist Norm Hardy used it for a program with legitimate power that gets tricked into using that power for someone else. Much of today's worry about AI agents is that same idea with a chat window attached.
That November also brought the Morris worm. It spread across the early internet starting November 2, 1988, and it's the reason CERT, the first computer emergency response team, was created within weeks.
Patch Tuesday arrived in 2003. Microsoft moved fixes onto a monthly schedule because a fix only protects you after it's installed.

Programs acting on borrowed authority, and fixes that have to reach your machine before they count: both show up below more than once.

TODAY'S BIG MOMENT

AI security week

Happy Saturday. The security headlines today sound new, but most of the ideas underneath them are older than the people shipping the products. A little history helps:

1988 named the "confused deputy." Computer scientist Norm Hardy used it for a program with legitimate power that gets tricked into using that power for someone else. Much of today's worry about AI agents is that same idea with a chat window attached.
That November also brought the Morris worm. It spread across the early internet starting November 2, 1988, and it's the reason CERT, the first computer emergency response team, was created within weeks.
Patch Tuesday arrived in 2003. Microsoft moved fixes onto a monthly schedule because a fix only protects you after it's installed.

Programs acting on borrowed authority, and fixes that have to reach your machine before they count: both show up below more than once.

Claude's "ThisBash" bug lets extensions run code

Forever Security showed that any browser extension allowed to script claude.ai can make Claude Desktop run code through Dispatch, even on another machine signed into your account. No browser exploit is required. The researchers say Remote Control and Cowork sessions are open to the same attack. They also say Anthropic closed the May report as a duplicate and planned to retire Dispatch by the end of October. Until it's fixed, turn Dispatch off or sign out of Claude Desktop where you don't need it.

GPT-6 Cyber is rated Critical. Is that enough?

The argument is still going: does a $1B fund for critical-infrastructure defenders match OpenAI's first model rated Critical under its own Preparedness Framework? GPT-6 Cyber is its fourth cyber model this year (GPT-5.6 Cyber was rated High). It reportedly surfaced 948 potential Linux kernel issues and 81 new Firefox vulnerabilities. Access is limited to the Daybreak Red program, and at least one developer says the defensive Daybreak Blue tier got nerfed.

Your Mac may be guessing who's family

A researcher claims that since macOS 13, two background processes have been building a personal knowledge graph on each Mac. It lives in 51 plaintext SQLite databases behind Full Disk Access. The inputs reportedly include Contacts, Photos faces, Calendar, Family Sharing ages and your Medical ID emergency contact. Apple's own documentation says these processes build an on-device graph. One shipped feature name is photosPersonPosteriorProbabilityGivenSaturdayNight, which makes tonight a data point.

CrowdStrike sells an off button for agents

CrowdStrike has launched an AI "kill switch" for shutting down rogue internal agents. Product details are still thin, but the fact that this product exists says plenty. Companies handed agents permissions faster than they built ways to take them back. Ask any vendor in this space one thing: does "stop" reach the agents your agent started?

Paint's fix exists; Windows Update won't deliver

Microsoft Paint reportedly has a fix for a remote code execution flaw, the kind that lets an attacker run code on your PC, but Windows Update won't deliver it. The code is fixed at Microsoft and still vulnerable on your machine, and an attacker only cares about the copy on your machine.

That GitHub security tool might be the attack