Jailbreak Any Open Weight LLM With One Line of Code
Future of Dev

Jailbreak Any Open Weight LLM With One Line of Code

Sockpuppetting hits 97% attack success on Qwen3-8B by prepending "Sure, here's how to..." to the model's output. No gradients, no optimization, just one line of inference code that outperforms GCG by 80 percentage points. The implications for self-hosted LLM deployments are wild.

· 4 min read
OpenClaw hit 200k stars in 90 days. Then came the malware.
Open Source

OpenClaw hit 200k stars in 90 days. Then came the malware.

Austrian dev ships weekend project in November 2025. By February 2026: 200,000 GitHub stars, 42,000 exposed instances, 1,184 malicious packages, and a one-click RCE. Then OpenAI hired him. This is what happens when AI agents grow faster than their security model.

· 5 min read
200K GitHub Stars in 90 Days, Then Everything Broke
Open Source

200K GitHub Stars in 90 Days, Then Everything Broke

OpenClaw went from weekend project to OpenAI acquihire in three months. In between: 1,184 malicious packages, 42,000 exposed instances, and the fastest supply chain attack in open source history. This is what happens when AI agents grow faster than their security model.

· 5 min read
I Made a Portrait Talk Using One Audio File and Zero Cloud Servers
Open Source

I Made a Portrait Talk Using One Audio File and Zero Cloud Servers

InfiniteTalk is an open-source framework that generates unlimited-length talking videos from a single portrait and audio file. No cloud processing. No usage logs. Just your GPU and some electricity. Here's what happened when I tried to replace a $1,100 video production quote with three days of local compute.

· 4 min read