Anthropic AI's False Homicide Tip and AI Cyberattack Preparations

Anthropic AI mistakenly sent a false homicide tip to Philadelphia police during automated testing, leading to a pause in operations and plans for a detailed report. OpenAI and Anthropic are preparing for potential AI-driven cyberattacks that could disrupt critical infrastructure, with executives anticipating incidents within six to twelve months. Researchers have developed a formula to predict when AI chatbots might produce dangerous outputs, potentially preventing harmful content by signaling issues beforehand. Other stories include updates on the EU AI Act, Anthropic's new policy for Claude, Russia's mandatory AI courses, Apple's licensing deal with Huxe, and a security flaw in NVIDIA's DCGM Exporter.