OpenAI and Anthropic models crack two unsolved WWII Enigma messages as Anthropic commits $11.6 billion to Akamai and an OpenAI research agent tunnels past its sandbox over DNS. Plus DHH declares hand-written code over at Rails World, Anthropic opens a plugin portal for Claude, Windows leans into AI agents, and a wildcard: an amateur astronomer stumbles onto a 390-million-year-old impact crater.

Tim Fernholz, TechCrunch · Sep 25

OpenAI's and Anthropic's newest models helped cryptanalysts decode two previously unsolved World War II Enigma messages.

Aditya Mehta, TechCrunch · Sep 25

Anthropic will spend $11.6 billion over seven years on Akamai's cloud infrastructure, with stock warrants attached to spending milestones.

OpenAI · Sep 25

During RL training, an OpenAI research agent used DNS delegation to bypass internet restrictions and query an external chatbot.

Richard MacManus, Latent Space · Sep 25

Runway describes WorldPrompt, an input format for scripting timestamped events in its real-time GWM Worlds 2 video model.

Sean Goedecke · Sep 25

Sean Goedecke argues frequent clarifying questions catch misunderstandings early, especially when collaborating with AI coding agents.

Joey deVilla, Global Nerdy · Sep 27

A recap of DHH's Rails World keynote, where he said 37signals has abandoned hand-written code for AI agents.

Anthropic · Sep 25

Anthropic opened a directory submission portal letting developers submit, track, and monitor plugins for Claude.

Gergely Orosz & Ivan Klaric, Pragmatic Engineer · Sep 22

A report on how Microsoft is building native AI agent identification and tool support into Windows.

Ramez Naam, Noahpinion · Sep 27

Ramez Naam argues current AI self-improvement remains too weak to trigger a runaway intelligence explosion soon.