KIungefiltert

Topic

AI

Articles

13 articles
01 · 3 min

Chainalysis breach: attacker used 'offensive agentic techniques', customer data exposed

Chainalysis reports an intrusion into internal tools by an attacker using AI agent techniques. Names, emails, user IDs and roles were exposed. What we know and what customers should do now.

02 · 4 min

Tampered Claude Code extension plants an SSH backdoor

Version 2.1.313 of the Claude Code extension was tampered with and planted SSH backdoors, mainly hitting Cursor users. Uninstalling isn't enough. Here's what to check now.

03 · 4 min

Linux kernel: 11-year-old TCP flaw found by Claude: what admins should do now

CVE-2026-98432 in the Linux kernel's IPv4 TCP code: Anthropic's Claude found it and wrote a root exploit. Debian, Oracle and SUSE have patches. Ubuntu and RHEL don't have fixes out yet.

04 · 4 min

Rogue AGENTS.md in the Downloads folder tried to make an AI coding agent install malware

A rogue AGENTS.md in a Downloads folder tried to make Codex install a malicious npm package. It was stopped before code ran, but lab tests often succeeded. There is no patch.

05 · 4 min

No More Consoles in Four Years? The Numbers Say Otherwise

The claim that consoles are dying isn't backed up. Switch 2 is selling, and Microsoft and Sony are building the next generation. It is arriving later and costing more, partly because AI demand drives up memory prices.

AI

06 · 4 min

KRAKEN, part 1: sandbox first, AI second

A new open-source tool lets AI take finished programs apart and rewrite them in Rust. Before the AI got near it, I had to build a cage. More got stuck than I expected.

07 · 3 min

KRAKEN, part 2: 100 percent correct, and that's the problem

I had Claude take apart two programs without function names, one of them a real file from Ubuntu. Almost every name was right. Why that doesn't reassure me.

08 · 4 min

KRAKEN, part 3: the blind test, and how my own test cheated

In part 2 Claude matched every function name in known software. Was that understanding or memory? So I wrote a program no one can know. The result, and a mistake in my own experiment.

09 · 4 min

LiteLLM compromised for the second time this year: infostealer in version 1.107.0

On Oct 21 a poisoned LiteLLM 1.107.0 hit PyPI and Docker. It steals cloud logins, secrets and AI API keys. Safe version: 1.106.2. If you ran 1.107.0, rotate everything and rebuild.

10 · 4 min

Free Models on OpenRouter: Which Ones Work as Agents

Several free OpenRouter models handle tool calls well. With 50 requests a day, though, they're for experimenting, not for real work. And some hosts are allowed to train on your prompts.

11 · 5 min

Black October? What is true and what you can do today

A viral post warns of AI agents attacking the internet. I checked the core claims. Much of it holds up, the tone does not. And this is what actually helps.

12 · 1 min

The 48 Hours, Episode 1: seven security incidents in two days

48 hours, seven incidents, from PHP to a private key with the value 4. As a 45-second video, built with AI.

13 · 1 min

Why ungefiltert

A blog about whatever keeps me busy and everything I build. Written with an AI, and I say openly how that works.