An AI voice clone drains a father's savings, Tesla limits staff AI spending while exempting its own tools, Anthropic talks custom silicon with Samsung, a live open-source world monitor tracks ships and cables, and a clever trick cuts Claude Code's token bills by more than half.

The Federal Reserve predicts AI will add jobs rather than destroy them, while Amazon designs its own AI silicon for Echo and Fire TV. We also cover Gemini Spark on Mac, rising gadget prices, Moonwalkers that turn walking into a sprint, Anthropic's life-sciences hackathon, and Palantir's CEO calling out AI pricing practices.

Anthropic launches Claude Sonnet 5 and a new Claude Science app, while Google ships Gemini Nano Banana 2 Lite, Omni Flash, and NotebookLM explainer videos. Plus: OpenAI's GeneBench-Pro benchmark, Etched's AI inference chip, HeyGen's prompt playbook, and Robert Greene on irreplaceable skills.

Meta demos a non-invasive brain decoder that writes full sentences from brainwaves, Cursor ships an iOS agent console, and AI-generated art keeps beating humans in contests. Plus, Mark Zuckerberg makes the case for personal AIs instead of one all-powerful model, Google reportedly caps Meta's Gemini compute, and FluidVoice brings offline speech control to macOS.

AI news roundup for 2026-06-28. TAAFT — Five Eyes AI Warning, Gemini Computer Use, Daybreak Patching. From TAAFT, your daily AI briefing.

Google AI Studio redesigns UI with one click, the US un-bans Claude Mythos, Apple Vision Pro chief leaves for OpenAI, LOOI robot uses your phone as the brain, and RuView turns WiFi into presence detection.

OpenAI reveals its own inference chip built in nine months, an AI law firm wins a real court case, Claude breaches NSA systems in hours, Gemini 3.5 Flash gains computer control, and AI reads 2,000-year-old scrolls. Your AI Daily News for June 25, 2026.

OpenAI ships its own inference chip, code-named Jalapeño, an AI-driven design pipeline puts a nuclear test flight vehicle in the air in months, and Google's Gemini 3.5 Flash gains direct computer control. Plus, a model that reportedly broke into NSA systems in hours, an AI law firm that won a real trial, and machine learning that recovered two-thousand-year-old unread text.

The Five Eyes alliance warns AI threats are months away, ElevenLabs launches Ads Engine for 50+ language localization, OpenAI ships Daybreak for automated patching, and an AI agent nukes a city in a simulation — then loses anyway. All this and more on AI Daily News for June 22, 2026.

A single ad reworked into fifty languages, a rare joint warning from the world's top signals agencies, and a Codex-based system that ships vulnerability fixes at machine speed. Plus, an AI civilization sim escalates to nukes, a survey on consumer AI control, a glowing AI desk dock, and OpenAI's blueprint for work that outlives one prompt.