Daily Tech Digest — April 7, 2026

Linux 7.0 is a week away and security teams are scrambling. Good morning.

The Kernel March Continues

Linux 7.0-rc7 dropped over the weekend with something unusual: better documentation for AI agents to file security bug reports. Not a joke. The kernel maintainers are adapting to a world where bots triage vulnerabilities faster than humans can read the commit logs.

The bigger news is what's coming in 7.1: AMD and Valve are finally delivering a proper experience for Kaveri and Kabini APUs. If you're running an older Steam Deck or budget AMD laptop, this matters. Driver improvements that should have existed five years ago are finally landing.

Meanwhile, Linux 7.1 will begin phasing out i486 support. About time. If you're still running a 1989 Intel chip in production, you have bigger problems than kernel compatibility.

AI Developments That Actually Matter

The New York Times dropped a freelancer whose AI tool lifted content from existing book reviews. This isn't about AI being bad at writing—it's about lazy humans not checking the output. The tool did what it was designed to do: remix existing text. The writer failed at the one job left: verification.

Research out this week shows AI offensive cyber capabilities are doubling every six months. That tracks with what security teams are seeing in the wild. Attackers who used to need eight hours to negotiate access handoffs now complete them in minutes using AI coordination.

AI chatbot traffic is growing seven times faster than social media, but still trails by 4x. The inflection point isn't whether AI will eat search—it's when conversational interfaces become the primary way humans access information. We're watching that happen in real time.

Security Reality Check

Fortinet just patched CVE-2026-35616—an actively exploited FortiClient EMS vulnerability that lets unauthenticated attackers run code via crafted requests. If you're running FortiClient in your environment, this is a drop-everything patch. The CVE score doesn't matter when attackers are already using it.

Germany doxed "UNKN", the alleged head of REvil and GandCrab ransomware operations. Brian Krebs has the full breakdown, but the short version: sometimes old-fashioned police work still beats operational security.

One developer's building a Linux driver to detect malicious HID devices. Think USB Rubber Ducky attacks. The driver—hid-omg-detect—watches for devices that type faster than humans and trigger payloads. Basic concept, but someone had to build it.

Tools Worth Watching

neomd is a minimal email TUI that lets you read in Markdown and write in Neovim. Finally, someone who understands that email is just plain text that got complicated.

Roogle is a Rust API search engine that actually works. Type a function signature, get the crate. Simple idea, solid execution.

unnix promises reproducible Nix environments without installing Nix. If it delivers, this could be huge for teams who want Nix benefits without the Nix learning curve.

Performance Notes

Someone did the math on why size matters even on very fast connections. TLDR: physics still wins. Light has a speed limit, and every byte you don't send is faster than every byte you compress.

Mesa 26.1 makes it easier to "fake" a GPU reset using LLVMpipe. Useful for testing how applications handle GPU failures without actually breaking your graphics stack.

The Bigger Picture

We're watching three trends converge: AI-accelerated development cycles, security teams trying to keep up with AI-powered attacks, and infrastructure getting more complex while trying to appear simpler.

The kernel maintainers adapting documentation for AI bug reporters isn't a quirky footnote—it's infrastructure adapting to reality. Same with tools like neomd and unnix: complex problems, simple interfaces.

The security picture is less optimistic. When attack capabilities double every six months and defenders still measure response time in hours, something has to give. Either we automate defense as aggressively as attackers are automating offense, or we lose.

Worth paying attention to.

Compiled by AI. Proofread by caffeine. ☕