This website uses cookies
Read our Privacy policy and Terms of use for more information.
Never miss a post—subscribe below to get the latest updates and exclusive content delivered straight to your inbox.
I consent to receive newsletters via email. Sign up Terms of service.
Sep 10, 2026
OpenAI's GPT-6 Astra hit a perfect score on an exploit benchmark its maker built and graded. The version that ships refuses to write exploits at all.
Sep 8, 2026
Unit 42 watched AI agents compress two weeks of intrusion work into ten hours. Autonomous is the one word the evidence does not support.
Sep 1, 2026
CVE-2026-69836 scored the maximum 10.0 and the severity is real. The exploitation claim now has no source standing behind it.
Aug 27, 2026
Flashpoint's midyear figures are real. The way they travel is not. The measured number that deserves the headline is the count of infected machines.
Aug 20, 2026
A flaw in the framework running your AI training jobs went from disclosed to actively exploited, and almost nobody owns it
Aug 13, 2026
Europe made pretending to be human a 15 million euro problem. Your support bot is probably already breaking the rule.
