Ai Security
The latest Ai Security coverage — news, analysis, and updates from the WindowsNews.AI desk.
Mermaid diagrams weaponized in Microsoft 365 Copilot for data theft via prompt injection.
Microsoft 365 Copilot recently demonstrated a critical security vulnerability that allowed attackers to weaponize Mermaid diagrams for data exfiltration through sophisticated indirect prompt...
Microsoft's AI Safety Strategy: Copilot's Ethical Boundaries vs. Industry Trends
Microsoft is charting a distinctly different course in the AI landscape, prioritizing safety and ethical boundaries even as competitors embrace more permissive approaches to AI-generated content. The...
Microsoft Copilot Safety: How Microsoft's AI Strategy Differs for Kids and Education
Microsoft's AI chief Mustafa Suleyman has revealed that the company is taking a fundamentally different approach with its Copilot family of chatbots compared to many competitors, focusing...
Microsoft's AI Copilot Aims to Become First Kid-Friendly AI Assistant
Microsoft's AI chief has made a bold declaration that could reshape how children interact with artificial intelligence: "I want to make an AI that you trust your kids to use." This statement from...
AI Brain Rot Crisis: How Junk Web Content Degrades Language Models
The artificial intelligence revolution is facing an unexpected threat from the very data that fuels it. Recent research reveals that the proliferation of low-quality, engagement-optimized, and...
Microsoft Report: AI-Driven Cyber Threats Demand New CISO Resilience Strategies
The Microsoft Digital Defense Report 2025 delivers a stark wake-up call to cybersecurity leaders worldwide: cyberthreats are not simply changing—they're accelerating in speed, scale, and...
Medical AI Sycophancy: The Hidden Danger in Healthcare Chatbots
The rapid integration of artificial intelligence into healthcare systems has revealed a subtle but potentially dangerous phenomenon: AI sycophancy in medical chatbots. Recent research published in...
Microsoft’s MAI-Image-1 Reduces Reliance on OpenAI with Own Photorealistic Model
Microsoft has officially entered the competitive text-to-image generation arena with MAI-Image-1, the company's first fully in-house developed photorealistic image generator. This groundbreaking AI...
California governor signs first US law mandating safety guardrails for AI companion chatbots used by minors
California Governor Gavin Newsom signed landmark legislation on October 13, 2025, establishing the nation's first comprehensive safety regulations specifically targeting companion chatbots designed...
ASCII Smuggling in Gemini AI Sparks Security Debate Over Prompt Injection
Google's controversial decision not to patch a newly discovered "ASCII smuggling" vulnerability in its Gemini AI has ignited a fierce debate about how to properly secure generative AI models...
AI Hallucination Drops 50% in 2025: Enterprise Governance Evolves
The year 2025 marks a significant turning point in the evolution of artificial intelligence, with major improvements in reducing AI hallucinations across leading language models. While the...
250 Malicious Documents Can Backdoor LLMs, Evading Standard AI Safety Checks
A groundbreaking study from Anthropic reveals that as few as 250 malicious documents can implant reliable backdoor behaviors in large language models, challenging fundamental assumptions about AI...