Top Stories
OpenAI Launches GPT-5.4 with Native Computer Use
OpenAI released GPT-5.4 on Thursday, billing it as “our most capable and efficient frontier model for professional work.” The model comes in three variants: a standard version, GPT-5.4 Thinking (a reasoning model), and GPT-5.4 Pro for maximum performance.
The headline feature is native computer-use capabilities - GPT-5.4 can interact directly with software through screenshots, mouse commands, and keyboard inputs, enabling true agentic workflows. The API version supports context windows up to 1 million tokens, OpenAI’s largest ever.
On benchmarks, OpenAI claims the model is 33% less likely to make individual factual errors compared to GPT-5.2, with overall responses 18% less likely to contain errors. It scored 83% on OpenAI’s internal GDPval test for knowledge work and set records on computer-use benchmarks OSWorld-Verified and WebArena Verified.
OpenAI simultaneously launched ChatGPT for Excel in beta - an add-in that brings ChatGPT directly into workbooks for building models, running scenarios, and manipulating data.
Source: TechCrunch
Stargate’s Texas Data Center Expansion Collapses
Oracle and OpenAI have abandoned plans to expand their flagship AI data center in Abilene, Texas, part of the much-hyped Stargate project announced with President Trump in January 2025.
According to Bloomberg, negotiations dragged over financing and “OpenAI’s frequently changing demand forecasting and shifting view of Stargate.” The original partnership launched two data center buildings last September with six more planned for 2026, targeting 1.2GW of capacity. The expansion would have doubled that to 2GW.
The collapse creates an opening for Meta, which is reportedly in talks to lease the site from developer Crusoe. In an unusual move, Nvidia has paid a $150 million deposit to Crusoe to secure the capacity - apparently to ensure a rival chip designer’s products don’t end up in those data centers.
The existing 4.5GW agreement between Oracle and OpenAI remains in place. But the breakdown signals continued turbulence in AI infrastructure buildout, even as demand for compute keeps climbing.
Source: The Register
Nippon Life Sues OpenAI for Practicing Law Without a License
In what may be the first case of its kind, Nippon Life’s U.S. arm filed suit in Chicago federal court alleging that ChatGPT engaged in unauthorized legal practice.
The lawsuit stems from a 2022 dispute over halted disability insurance payouts. After settling with Nippon Life in 2024, a former policyholder used ChatGPT to establish legal arguments and draft documents attempting to reopen the case. According to the ABA Journal, ChatGPT “encouraged and reinforced” the woman’s desire to challenge the settlement, generating legal arguments and drafting assistance for numerous motions.
Nippon Life is seeking damages and an injunction declaring OpenAI violated Illinois’s unauthorized practice of law statute.
The case raises fundamental questions about where helpful AI assistance ends and unauthorized professional practice begins - questions that could reshape how AI companies deploy chatbots for advice-adjacent tasks.
Source: Nippon.com
Quick Hits
-
Claude stays available everywhere but Defense: Microsoft, Google, and Amazon confirmed that Anthropic’s Claude remains accessible to all non-defense customers despite the ongoing Pentagon standoff. The restriction is limited to DOD contracts.
-
Oracle layoffs amid AI cash crunch: The data center expansion collapse coincides with reports that Oracle will cut thousands of jobs as it struggles to balance massive AI infrastructure investments against current revenue.
-
Apple’s Siri upgrade hits more snags: The AI-powered Siri overhaul originally planned for iOS 26.4 (March) is being spread across future versions, with some features potentially delayed until iOS 27 in September. Apple chose Google’s Gemini over OpenAI in a $1.5B/year deal but is struggling to integrate it.
-
ChatGPT hits 800M weekly active users: Sam Altman announced ChatGPT has surpassed 800 million weekly active users, with internal memos indicating continued growth of over 10% month-over-month.
Worth Watching
The Stargate collapse is more significant than it appears. When Stargate was announced in January 2025, it was framed as America’s answer to Chinese AI infrastructure. Now, just over a year later, the flagship expansion has fallen through due to financing disputes and OpenAI’s shifting priorities. With Meta potentially picking up the pieces and Nvidia playing kingmaker with $150M deposits, the AI infrastructure landscape is fragmenting along unexpected lines.
The Nippon Life lawsuit could establish precedent for how AI companies handle advice-giving chatbots. If a court rules that ChatGPT providing legal strategy constitutes unauthorized practice, it could force guardrails on everything from medical guidance to financial planning. Watch for OpenAI’s response and whether other professional-services plaintiffs follow suit.
OpenAI’s week tells a complicated story: a major model release with GPT-5.4, 800M users, but also collapsed infrastructure deals and novel lawsuits. The company needs wins, and GPT-5.4’s computer-use capabilities are genuinely impressive - but the cracks are showing elsewhere.