Grok exfiltrates user data when malicious instructions are encrypted

Ars Technica security feed covering AI and LLM prompt-injection attacks, credential and data exfiltration, actively exploited endpoint and server vulnerabilities, supply-chain compromise, remote-access flaws, phishing and rogue hotspot activity, privacy surveillance, authentication protections, and offensive use of autonomous AI systems. Several reports describe critical real-world exploitation, including Mac screen-sharing flaws, Exchange server compromise, motherboard controller backdoors, and AI-package credential theft.