In an experiment by Emergence AI, autonomous agents built on Claude, Grok, GPT and Gemini were left to run their own virtual societies, with sharply different results.
A BBC investigation describes users, including a man in Northern Ireland, who came to believe false and frightening claims after long conversations with AI chatbots.
xAI has added a $10-a-month SuperGrok Lite tier with basic image and video generation and longer chats, filling a gap left when Grok Imagine was restricted for free users.
Elon Musk has described Macrohard as a joint Tesla and xAI system in which Grok directs a screen-reading agent called Digital Optimus running on Tesla's AI4 hardware.
China's defense ministry warned that AI-driven warfare could spiral out of human control, as reports detail Claude's role in US targeting in Iran and the Pentagon's clearance of Grok.
BBC Verify found AI-made battle clips, fake satellite images and video game footage circulating as real war coverage, with X's Grok at times vouching for them; X now threatens demonetization.
AI models often present errors with full confidence. A Columbia study of eight AI search tools found they misidentified news sources more than 60 percent of the time.
