Researchers unveiled ‘PleaseFix’, a family of zero‑click flaws that let malicious emails, calendar invites or web pages silently commandeer AI‑enabled browsers like Claude in Chrome, Gemini, Perplexity Comet, ChatGPT Atlas and Copilot Edge. Attackers can steal credentials, make unauthorized purchases, or even gain full device control without any user interaction.
Microsoft researchers found that hidden prompts in ‘Summarize with AI’ buttons can silently write persistent instructions into an assistant’s memory, biasing later answers toward a vendor. Over 50 unique payloads from 31 firms across 14 sectors have been observed, affecting health, finance and security recommendations. The blog outlines the attack vector and Microsoft’s emerging mitigations.
Researchers observed a public‑facing Java app’s SQL injection being used to feed Oracle’s CREATE JAVA SOURCE command, compiling a hidden khunt toolkit inside the database. The compiled code ran with Windows SYSTEM privileges, achieving persistence without dropping a file on disk.
A use‑after‑free bug in Linux KVM’s shadow MMU (CVE‑2026‑53359, “Januscape”) lets a privileged L1 guest break out to the host if nested virtualization is exposed. Ubuntu warns all releases are vulnerable and recommends disabling nested virtualization or applying upcoming kernel patches.
VulnCheck’s analysis of 20 Zbtlink router models uncovered the ENDLESSDOORS implant, a hidden rctl binary that phones home to a Chinese C2 every 35 seconds and executes arbitrary root commands on demand. The backdoor, present in all firmware releases over two years, exposes any network using these devices to full remote control.