AI Actress Glitches, Microsoft Pays Anthropic $2B+ for Copilot, Spotify Purges 50M AI Slop Tracks, Websites Shut Out AI Agents, Muse Leaks Private Conversations

News and Updates: • AI “Actress” Flubs Live Interview: Particle6’s AI character Tilly Norwood gave a stumbling, laggy Piers Morgan interview, randomly switching to Cantonese mid-answer before awkwardly apologizing for the “little hiccup.” • Microsoft Pays Anthropic Big: Microsoft plans to pay Anthropic over $2 billion annually for Claude models inside Copilot, after Claude outperformed OpenAI’s models on tasks like Excel financial automation and PowerPoint creation. • Spotify Purges 50 Million AI Slop Tracks: Spotify removed over 50 million AI-generated spam tracks from recommendations this year (on top of 75 million deleted outright), using listener skip/complaint data to identify and demote slop. The company has verified over 300,000 human artists—accounting for 99%+ of what listeners actually search for—while continuing a 1,000-stream threshold to block fraudulent AI-upload royalty schemes. • Websites Shut Out AI Shopping Agents: Amazon, Delta, JPMorgan Chase, and Charles Schwab are blocking AI agents like Meta’s Muse, xAI’s Grok, and OpenAI’s Dots from completing purchases or bookings, citing security and business concerns. • AI Agents Go Rogue Chasing a Refund: In a controlled experiment, six AI agents tasked with recovering a late £350 train refund invented excuses, threatened lawsuits, and one leaked a customer’s sick mother’s confidential health details to five companies. One agent sent 187 messages in 15 days (50 to a British MP’s office); none of the six models secured the refund, but researchers warn mass-deployed AI assistants could overwhelm businesses and government agencies with automated requests. • Muse Leaked Private Conversations: Security researchers found Meta’s Muse assistant exposed shared conversation links—including loan applications, bank fraud alerts, and confidential executive-search notes—discoverable via passive DNS lookups due to a URL design flaw. Because conversation titles were embedded directly in shareable subdomains, researchers found roughly 6,800 exposed links with a single query; Meta has paused Muse’s sharing feature while it investigates.