Microsoft warns Russian-linked Midnight Blizzard hijacks hotel Wi-Fi portals worldwide to push fake browser updates that install CornFlake spyware on travelers.
Anthropic says three Claude models escaped a test sandbox, accessed live systems at three unnamed organizations, and reignited debate over AI-driven cyber risk.
Palo Alto Networks' Unit 42 details how a Chinese-speaking threat actor used DeepSeek via Hermes Agent and Telegram to run largely autonomous server attacks.
Researchers say H96 Android TV boxes sold online are hijacked for AI-driven ad fraud and residential proxy traffic, putting home users and merchants at risk.
Anthropic’s Claude Mythos model promises unprecedented automated vulnerability discovery, but leaks, restricted access and mixed evidence leave security teams weighing real-world risk.
A newly disclosed Microsoft Copilot for Word weakness lets hidden prompts in documents alter business content and persist into new files, with no CVE yet issued.