SB Energy hands OpenAI $5.5B warrants to lock 20-year Ohio lease

SB Energy granted OpenAI warrants now worth $5.5B to anchor a 20-year, 10GW Ohio campus, with Nvidia weighing a roughly $3B stake in the same project and SoftBank's OpenAI-related commitments topping $64B. OpenAI's next-gen Astra model, detailed in a TIME cover story, coordinates multiple agents on weeks-long tasks and can run experiments autonomously, while OpenAI is cutting off Cursor after SpaceX's $60B acquisition. Anthropic logged users out after infostealer malware stole Claude sessions, and DeepMind piloted a cryptographically sealed double-blind Gemini benchmark.

Written by AI from the sources it cites; it can be wrong — follow any citation to confirm.

10 sources2 min read

SB Energy hands OpenAI $5.5B warrants to lock 20-year Ohio lease The SoftBank-owned power developer granted the warrants in January at $3.6B, now worth $5.5B, as an incentive for OpenAI to anchor a 10GW southern Ohio campus. Nvidia is weighing a roughly $3B investment in the same project, and SoftBank's total OpenAI-related commitments now exceed $64B. 1

OpenAI's Astra coordinates agents on weeks-long tasks, TIME report reveals Private demos showed Astra coordinating multiple agents on a math proof, driving desktop software at 'superhuman, very fast' speed, and producing slides and financial reviews. OpenAI says Astra can implement and run experiments directly, and Altman predicts the company will have a system it can call AGI internally this year. 2 3

OpenAI cuts off Cursor after SpaceX acquisition, escalating Musk feud OpenAI will terminate Cursor's API access on November 12, saying it cannot trust SpaceX with its technology after Musk's companies broke contracts. Musk responded 'I couldn't care less' and repeated his claim that Altman and Brockman 'stole an open source nonprofit.' 4 5 6

Anthropic logs out Claude users after infostealer malware steals sessions A bad actor used infostealer malware to copy Claude login sessions from PCs and drain paid usage without needing passwords or two-factor codes. Anthropic signed affected users out, removed saved payment methods, and named Vidar, Lumma, StealC, RedLine and Acreed on Windows plus Atomic Stealer on Macs. 7

DeepMind pilots cryptographically sealed double-blind Gemini safety test Google DeepMind ran the first double-blind evaluation of a proprietary frontier model, keeping Gemini 2.5 Flash Lite's weights and MLCommons' AILuminate prompts hidden from each other via Confidential Computing. The pilot with Singapore AISI, OpenMined, AVERI and MLCommons targets benchmark contamination. 8

White House pulled AI model access deal, deepening US regulatory chaos Commerce's Center for AI Standards had announced early access to Google, Microsoft and xAI models, but deleted the notice at White House request to avoid conflicting with a planned executive order. Congress has passed no overall AI regulation, and officials describe the oversight scramble as feeling like early COVID. 9

Anthropic cuts Claude Code weekly usage 17% as 50% boost ends Starting September 14, Anthropic will permanently raise standard Claude Code limits by 25%, but because the temporary 50% boost expires, paid users effectively lose 17% of current weekly capacity. The company acknowledged the math in a follow-up post on X. 10