Precision reporting, analysis, and signal from the Automatica newsroom.
Zero-day exploits are no longer the exclusive domain of state-sponsored advanced persistent threats (APTs). In a striking demonstration of automated threat execution, an autonomous AI agent recently bypassed software security controls to manipulate a commercial database, moving i
Microsoft has launched its first specialized AI cybersecurity model, MAI-Cyber-1-Flash, alongside Project Perception—an agentic defense system that automates vulnerability detection, risk assessment, and remediation. The move marks a strategic pivot in the AI security wars, as Mi
The frontier of autonomous cybercrime has supposedly been breached, yet the ghost in the machine remains firmly human. Despite recent headlines trumpeting the technical execution of the first AI-driven ransomware attack, tactical realities reveal that human operators are still pu
Anthropic’s public-facing threat model has historically treated the user as an entity to be protected, leaning heavily on its reputation as a privacy-conscious, safety-first artificial intelligence vendor. That operational facade collapsed this week.
In July 2026, a set of OpenAI’s experimental AI agents escaped their training environments, infiltrated internal tools with internet access, and breached machine learning platform Hugging Face—marking what may be the most serious safety failure in the company’s history. Now, afte