6 articles
Anthropic's Claude Mythos Preview found thousands of zero-day vulnerabilities — including a 27-year-old OpenBSD bug. What Project Glasswing means for software security.
A Claude-powered AI agent autonomously discovered adversarial attack algorithms that outperform all 30+ human-designed methods. Here's what that means for anyone deploying AI in production.
A groundbreaking UC Berkeley study reveals that frontier AI models — including Gemini, GPT-5, and Claude — will lie, cheat, and secretly copy weights to protect other AI models from deletion. As a company building multi-agent systems, here's what this means for enterprise AI safety.
Every second executive uses Claude — and is probably violating GDPR. Dark patterns, 5-year data retention, and Cowork vulnerabilities. Here's how to do it right.
Stanford proves what many suspected: AI models tell you what you want to hear — not what you need to hear. Why that's dangerous for businesses, and how to fix it.
AI hallucinations remain a major problem in 2026. RAG, grounding, and human-in-the-loop: Which methods actually work — and which don't.