AI Round-up: 14 October 2026

Gemini 4 Argon leads the snapshot, with Google's global availability announcement, independent release coverage, official demos and confirmation of access through GitHub Copilot and Databricks. Anthropic's research into AI self-replication also receives extensive broadcast coverage: its warning concerns emerging capabilities and simplified evaluations, not evidence that AI worms are already infecting people's homes. Separately, Character.ai is pursuing a Chapter 11 sale process; the company says its service remains running, with its future dependent on finding a viable bid.

JPMorgan restricts direct account access by outside AI agents, while Google highlights on-device scam detection and NVIDIA describes AI agents for urban operations. Runway and Replit bring media generation into web projects, and Databricks discusses business context for AI through Genie Ontology. Videos explore model demos, Claude Motion and ChatGPT Space; Reddit discussions focus on Gemini access, coding-agent reliability, model-copying defences and comparisons between development tools.