Ep. 52 — OpenAI previews safety monitoring that keeps nothing while Anthropic keeps thirty days, Slack moves code review into the channel, and Binance hands agents the trade button

OpenAI previews safety monitoring that keeps nothing while Anthropic keeps thirty days, Slack moves code review into the channel, and Binance hands agents the trade button 00:48 — OpenAI previews Private Safety Processing — safety monitoring that keeps zero customer data — while Anthropic's covered models keep 30 days of prompts 02:17 — Slack launches Slack Code: tag a coding agent in a channel and it opens its own workspace with diffs, previews, and an audit log 03:47 — Binance opens spot and futures trading to AI agents — withdrawals blocked by default, and the exchange says it cannot see the agent's reasoning 05:17 — Security researchers lose access to OpenAI's Daybreak Blue cyber tier; OpenAI calls it a technical issue and asks them to re-verify 05:42 — Cognition CEO Scott Wu denies Bloomberg's report that SpaceX tried to buy the company — while Bloomberg says a new round is being discussed at $40 billion 06:06 — Meta ships a Mac app with system-wide dictation and screen reading — and pipes merchants' ad accounts into it 06:30 — Google puts test prep and a notebook inside Search — free practice quizzes for the SAT, MCAT, LSAT and more, with test-prep partners supplying the questions Daily Prompt with Archer & Iris.