Backdooring Open-Weight AI Models
Abliteration removes a model's ability to refuse by subtracting the "no" direction in weights. Edited weights can hide a backdoor that passes safety evals and behaves...
Abliteration removes a model's ability to refuse by subtracting the "no" direction in weights. Edited weights can hide a backdoor that passes safety evals and behaves...
The flaw is a remote code execution vulnerability in the PAN-OS management interface, tracked as CVE-2026-0322, with a CVSS score of 8.6, and attackers are actively...
A suspected advanced threat actor has been observed exploiting a security flaw in Fortinet FortiOS to deliver a previously unreported memory-resident ARM64 malware....
The China-nexus cyber espionage group Armory Castle, also known as UNC5221 or Warp Panda, exploited CVE-2026-102764 in late August 2026. The vulnerability has a CVSS...
F-Secure launches a victim-first resource. Built from decades of threat intelligence research, it explains scams from the victims' perspective, tracks emerging cyber...
A researcher demonstrated the leak on Vercel's live cloud sandboxes, and Vercel confirmed it. The vulnerability, tracked as CVE-2026-99999 and nicknamed PwnEscape,...
A threat actor uploaded malicious packages to the Arch User Repository that impersonated two popular helper tools, yay and paru. These fake packages run with root...
Today's update fixes more than 1,400 vulnerabilities in Windows and other software. This batch dwarfs September's record of nearly 1,000 fixes. One weakness is a zero...