Threats

A planted instruction file in a developer's Downloads folder attempted to make an AI coding agent install malware-linked npm packages, researchers found.

CVE-2026-99999, dubbed PwnEscape, lets code inside a VM read host memory. Patches are available for all supported Linux stable branches.

An unknown threat actor used the open-source AI pentesting tool ARTEX to breach South Korean banks. A separate actor, SCARLET LOOP, was found automating account

OpenAI disrupted a coordinated distillation campaign attempting to illicitly extract protected reasoning from its AI models, attributing core activity to Moonsh

DPRK-linked attackers are distributing Go-based malware via Terraform providers and Go modules, marking a novel supply chain distribution vector.

ShinyHunters breached rival ransomware gang Cl0p and threatened to expose victim data. Organizations that paid Cl0p may now face renewed extortion risks.

Anthropic's security document reveals Claude still struggles with CAPTCHAs, while unverified reports claim GPT-6 Astra solved all 48 levels of a CAPTCHA game.

A New Jersey court ordered radaris.com and over a dozen related domains transferred to Atlas Data Privacy Corp after Radaris repeatedly evaded litigation.

A September 23 webinar from BleepingComputer and Material Security examines real Google Workspace breaches and the response decisions that shaped their outcomes

Microsoft's September patch fixes a record 972 vulnerabilities, 112 of them critical. AI-powered vulnerability discovery is driving the surge.

Schneier's DEF CON talk on AI hacking drew over 100K YouTube views. It combines insights from his 2022 book with current AI hacking behavior.

A new cPanel flaw lets an authenticated hosting account achieve root-level code execution via SQL injection in EmailTrack. All supported versions are affected.