OpenAI's DevDay Pitch: Cloud Agents Your Whole Team Shares

OpenAI's DevDay sessions describe agents that live in the cloud and belong to a team instead of one person, and the open issue is who can edit, trigger and review them. Also today: a Senate bill that would make data centers pay for the grid, withdrawals and a credit dispute around OpenAI's math results, new attribution for the South Korean bank hack, the White House science summit, and a Brief. OpenAI DevDay: Codex Cloud session Cloud environments with filesystem snapshots, plus Tailscale and OIDC access to private infrastructure; it matters because a shared agent can now reach into your network. OpenAI DevDay: Dots and Spaces session A Dot is a persistent cloud agent with its own browser and file system, and the speaker lists team Dots as future work; the permissions model for several people sharing one is left open. OpenAI DevDay: OpenClaw session Shared team sessions make prompts and decisions visible alongside the code, which changes what review looks like. OpenAI DevDay: ChatGPT Work session Page instructions work like a shared config file and results queue for human review before publication. Techmeme on OpenAI's GPT-6 tiering GPT-6 Sol serves paid ChatGPT tiers and Luna serves Free and Go; the post says nothing more about the models. AI Engineer talk on Strands context handling Keep a pointer in agent state and fetch bulk data on demand, the opposite default from one shared workspace. Axios on the Senate data center bill Data centers would pay existing and incremental transmission costs, and new sites of 20 megawatts or more would bear incremental system costs; the bill is a proposal only. Washington Post via Techmeme on the Fed September meeting officials cited AI investment in discussing inflation. Hacker News discussion of OpenAI's withdrawn math results Three results reported withdrawn; the thread title is the extent of what we have. doomslide on X A prior-work allegation, treated here as an allegation. Susan Zhang on X Second post in the credit dispute. Axios on OpenAI's math results Why verifiability makes math the next field, with Stephen Wolfram's skepticism. CrowdStrike analysis via Techmeme The South Korean bank attacker is likely Chinese-speaking and financially motivated, and used LLMs and ARTEX. LocalLLaMA post on the attacker's model stack Names five tools and models; unverified. Forbes on ransomware and autonomous AI A contributor's argument that defenders need AI too. Axios on the Genesis summit More than $1B in industry commitments and a southeastern compute hub, as announced. Reuters via Techmeme on Biohub $300M from Meta, DeepMind and Isomorphic and $500M+ from the US for open biology datasets. Bloomberg via Techmeme on Isomorphic Labs Early talks at a valuation of at least $40B. Claude Haiku 5.5 announcement Anthropic's own description of the model. AINews on Haiku 5.5 Tiered pricing and the subagent positioning. Indian Express on Haiku 5.5 The third model in a month ahead of a planned IPO. WSJ via Techmeme on the fired researchers' letter Their ask on AI monitoring; one side of a firing dispute. Guardian column on Sam Altman Altman's remark on accepting some bad things for the technology's benefits, with the columnist's argument. WSJ via Techmeme on Broadcom financing More than $50B in financing for OpenAI's custom chip, reported from sources. FT via Techmeme on SemiAnalysis China 24 GW operating with 50 GW planned, versus 56 GW in the US. CNBC on Manus $500M+ raise after Beijing forced Meta to unwind its acquisition. Politico via Techmeme on Ecosia Dropping Mistral for open-source models, some Chinese. Forbes on Sarvam A new hire to sell sovereign AI, now carrying DeepSeek and Zhipu models.