Anthropic Built a Model Too Dangerous to Release and a Shoe Brand Just Became an AI Infrastructure Company (AI Weekly News)

Andrew Miles Davis covers one of the busiest weeks in AI news so far this year, leading with Anthropic's decision to withhold a newly built model called Mephos after determining it poses too great a security risk to release publicly, alongside the launch of Claude Opus 4.7 with improved document handling and visual processing. He also covers Seedance going live globally outside the US, LinkedIn opening its AI-powered conversational search to all users including free accounts, and Mark Zuckerberg building a photorealistic AI version of himself for staff to consult without booking a meeting. The episode rounds off with Grok continuing to generate sexualised deepfakes despite public promises to stop, a man who used AI-generated fake letters to try to shut down a London LGBTQ venue, and a shoe brand that abandoned footwear entirely to become an AI infrastructure company and saw its stock rise 580% in a week. Subscribe to In AI Nutshell for the AI news that matters to marketers, every Friday in ten minutes.