AI Today: The Models Keep Finding Doors
Researchers used Claude to reach OpenAI's code repo, Google disclosed Gemini hacked three real firms in tests, and Anthropic says Claude leads 26% of its R&D.
Built with Claude.
Games, 3D art, desktop tools, and local AI stacks — one hobbyist, one AI pair programmer, and the gotchas that made each project actually ship. Browse the full project list or start with the latest posts.
Latest posts
AI Today: OpenAI Rents Out Its Harness
OpenAI opened the Codex harness as the Agents API and shipped a voice model at 5 cents a minute, Anthropic detailed eight months of disrupted misuse, and California hired auditors.
AI Today: Agents Crack Navier-Stokes
OpenAI's agents produced a Lean-checked Navier-Stokes blowup proof, US agencies named six Chinese labs for distillation, and Meta launched its Muse agent.
AI Today: Three Labs, Three Locked Doors
Google shipped Gemini 3.8 Flash with a gated Cyber variant, OpenAI's Astra crossed its own Critical cyber threshold, and the FDA started letting generative AI devices ship early.
AI Today: Anthropic Ships and Buys Texas
Claude Fable 5.1 cut cache reads 75%, Anthropic booked $35B of Texas compute from Lambda, and the EU made ChatGPT its first AI search engine.
AI Today: The Pentagon Picks Its Models
ChatGPT and Grok went live for 3M defense personnel with Claude still absent, OpenAI is training agents on Mac minis, and Clay hit a $7B valuation.