
Whatever AI Safety Is, It’s Not This
WIRED’s Barrett says Trump’s AI self-police handshake is seat-belt theater — applause in the Oval Office isn’t a safety standard.
The file
Sourced AI briefs in publish order — newest first. Citeable originals live on the homepage under From Signal Desk.

WIRED’s Barrett says Trump’s AI self-police handshake is seat-belt theater — applause in the Oval Office isn’t a safety standard.

Time’s sourcing puts Grok in a pre-invasion Oval chat — and the Pentagon’s AI chief later said Grok Gov helped pick Iran War targets.

Judge Mehta tosses Chegg and Penske’s AI Overviews antitrust suits — unfair to publishers, he says, but not an illegal agreement under current law.

Gemini Live’s Guided Vision narrates the camera feed for low-vision users — useful accessibility AI with a clear “not a cane” warning.

Global Try On and Favorites land in ChatGPT shopping — OpenAI chasing Pinterest/Google fashion discovery after checkout flopped.

WSJ: three OpenAI safety researchers out for allegedly sharing sensitive info with an outside safety org — days after Times security culture reporting.

NVIDIA’s DOCA agent skills turn BlueField coding agents from guessers into checklist passers — 19% to 100% on a 65-prompt eval.

DIN Deploy open-sources C++ ONNX Runtime + TensorRT RTX samples — Whisper to FLUX.2 — aiming at real local apps, not notebook demos.

The Verge’s cut of Mehta’s dismissal: Chegg and Penske lose the AI Overviews antitrust fight — disclosure, the outlet is Penske-owned.

OpenAI’s essay argues genius needs institutional muscle — and that AI’s first gift may be boring execution, not a billion Einsteins.

AWS open-sources Strands Decider 2B — a small, local “pick among options” model in the Jev mold as decision models multiply.

Albertsons expands OpenAI work — ChatGPT Enterprise inside, Safeway shopping inside ChatGPT for millions of weekly grocery trips.