OpenAI

Recent news

• Oct 1, 2026

  Nvidia built the AI factory. Now it’s building the locks for the doors

  Nvidia has announced the Open Agent Safety Platform, a new open reference design for governing AI agents with OpenShell and Sentry.

  • OpenShell is an Apache 2.0-licensed runtime that moves policy enforcement into the Linux kernel, isolates each agent in its own sandbox, keeps credentials in a gateway outside the sandbox, and uses formal verification to prove boundaries before execution.
  • Sentry runs on BlueField-4 DPUs as an out-of-band security domain that can monitor requests, responses, and chain-of-thought reasoning and intervene in milliseconds; Nvidia says partners including Anthropic, Salesforce, SAP, SpaceXAI, Citi, and JPMorganChase are working with the platform, and Nvidia plans to move OpenShell to the Linux Foundation / CNCF later.
• Sep 29, 2026

  Trump Says Top Tech Firms Have Signed Accord to 'Self-Police' AI Development

  The White House has announced that President Donald Trump and a group of AI company leaders agreed to a voluntary accord on AI safety and self-policing during a meeting in Washington.

  • Trump said the accord will include internal and external reviews of AI technology, and that companies agreed to use self-regulation while the government relies on existing enforcement through the Department of Justice and the FBI.
  • The article says the meeting included leaders from Anthropic, OpenAI, Amazon, X, Nvidia, Google, Meta, and Microsoft; Trump also said executives would help address local opposition to data centers by supporting schools or reducing energy costs.
• Sep 29, 2026

  How data centers factor into DTE’s 20-year power plan

  DTE Energy’s long-range power plan is under review by Michigan regulators and includes two new natural gas plants, major renewable and storage additions, and capacity growth through 2046.

  • Main action: DTE proposes 700 MW gas plant in Milan Township and 1,400 MW plant in Handy Township, with costs of $2.5 billion and $3.9 billion respectively.
  • Background/details: The plan would grow capacity from 16 GW to 33 GW by 2046; DTE says the Monroe coal plant retirement drives the need, while critics say the new gas is mainly to serve Oracle/OpenAI and Google data center load.
  • The utility also proposes 15 GW renewables, 4.5 GW storage, 177 MW additional nuclear at Fermi 2 starting 2036, and 2.1 GW of natural gas generation; the Michigan Public Service Commission has 300 days for initial review.
• Sep 28, 2026 · Outsourcing Today

  Corporate AI spending is set to rise by 406% by 2026 compared to last year (analysis)

  This article is an analysis of Revolut Business transaction data on corporate AI spending, not a first-time corporate announcement.

  • Revolut Business AI Data analysis shows business spending on AI products rose 406% versus 2025, based on anonymized and aggregated transactions from 850,000+ business accounts globally.
  • Spending was concentrated in OpenAI, Anthropic, and Cursor; companies spent $416 vs $42 per user compared with consumers, while Romania’s mid-market segment saw the fastest growth at 440.2% YoY.
  • The analysis covers January 2023 to August 2026 and includes related infrastructure such as GPU-based cloud services and AI model delivery platforms monitored since January 2024.
• Sep 25, 2026

  Gleb Tsipursky: AI Infrastructure Needs Authority Budgets Before Agents Get the Keys

  Broadband Breakfast published an expert opinion arguing that AI agents should be constrained by authority budgets before they are widely deployed. It is commentary and analysis, not a first-time corporate announcement.

  • The piece argues for an authority budget for AI agents: limited delegated power, least privilege, expiring permissions, logging, monitoring, and a fast pause/kill mechanism for higher-risk actions.
  • It cites recent examples and safety positions, including Jacob Coxon leaving Anthropic, Evan Hubinger warning AI could kill all humans, OpenAI’s Hugging Face incident report, and Dario Amodei’s call for stronger regulation and third-party evaluators.
• Sep 25, 2026

  El-Sayed, Rogers face voter opposition to data centers in crucial Senate race

  Planet Detroit reports on the Michigan U.S. Senate race and its overlap with data center policy, campaign finances, and voter sentiment. The article references past and current positions and financial disclosures rather than announcing a new project.

  • Mike Rogers said in August that he supports a one-year moratorium on new data centers; Abdul El-Sayed released “terms of engagement” in January and supports state and local moratoriums.
  • The story cites an August EPIC-MRA poll showing 62% of likely Michigan voters oppose data center development and notes Rogers’ disclosure of $100,001 to $250,000 in Blackstone Real Estate Income Trust, which has 29% of its portfolio invested in data centers.
  • It also mentions the $16-billion Oracle and OpenAI data center under construction in Saline Township, the $1.25-billion University of Michigan computing facility in Ypsilanti Township, and a draft wastewater permit for Related Digital with public comments open through Oct. 23, 2026.
• Sep 21, 2026

  Datacentres: are we frogs in a digital pot?

  East Anglia Bylines has published an opinion article arguing that AI is worsening datacentre energy and resource use, while the broader growth of cloud computing predates AI.

  • The piece explains that cloud services run on datacentre hardware, and says repatriating systems is costly; one project reportedly saved over 60% by bringing systems back in house, but it took three years.
  • It cites AWS as generating well over half of Amazon’s operating income in 2024, notes AWS’s 28% share of the worldwide cloud infrastructure market, and says server prices are up by a third or more year-on-year as AI demand lifts costs.
• Sep 18, 2026

  Shapiro calls on Trump to regulate artificial intelligence development

  Pennsylvania Gov. Josh Shapiro called for stronger federal and international regulation of AI during remarks at the AI Horizons summit in Pittsburgh.

  • He said the U.S. should push China for “practical guardrails” and argued that industry leaders should not exclusively guide the regulatory process.
  • Shapiro also criticized the Trump administration’s hands-off approach and said the state recently signed an executive order requiring data center developers to get local approval and removing expedited permitting pathways.
  • The article is a news report/commentary from 90.5 WESA and includes reactions from industry and labor leaders at the summit, plus references to AI companies calling for slower development.
• Sep 17, 2026

  11 AI Bills Cleared Committee this Summer, But None Has Reached the House Floor

  Rep. Jay Obernolte said the bipartisan AI package he co-authored could be combined into a single bill for a House floor vote in December.

  • 11 bills from the Great American AI Act were split and routed to relevant committees; the House Science, Space and Technology Committee cleared the science-related bills in July, while the FRONTIER Act remains with the Energy and Commerce Committee.
  • The package includes preemption of some state AI laws for three years, frontier-model disclosure and risk-management requirements, and potential federalization of frontier safety laws in California, New York, and Illinois; Obernolte and Lori Trahan introduced the original 269-page draft on June 4, and Obernolte spoke at Tarana Wireless’ Connecting Communities Summit in Milpitas, California.
• Sep 9, 2026

  If you lost your LLM tomorrow, would you still have a business?

  The article argues that European AI companies need sovereign fallback infrastructure and should plan for model outages, supplier changes, and regulatory restrictions rather than rely entirely on foreign API providers.

  • It cites a recent Anthropic incident in which the US Department of Commerce allegedly required access restrictions on Claude Fable 5 and Mythos 5, leading Anthropic to switch off the models for all customers and restore them 17 days later.
  • It says Europe should retain the ability to run models on internal infrastructure, use open-weight models such as Mistral, Gemma, DeepSeek, and Qwen, and it highlights constraints around funding, grid connections/data centres, and talent retention; it also references the EU’s 19 AI Factories and 13 Antennas attached to EuroHPC supercomputers.
• Sep 9, 2026

  🤖 La Machine #91: Mistral’s €3 Billion Mic Drop

  Mistral AI has announced a €3 billion funding round at a post-money valuation above €21 billion.

  • Samsung led the financing, which Mistral says is the largest private European tech equity round on record.
  • The funds are intended for frontier research, more compute, and Mistral’s own infrastructure; the article also says Mistral has nearly doubled its valuation in a year and is still positioning itself against OpenAI and Anthropic.
  • The piece is a newsletter-style news summary referencing a recent announcement rather than publishing Mistral’s original press release.
  • The article also mentions Mistral’s sovereignty pitch and says competitive models are coming “very soon.”
• Sep 3, 2026

  Uber To Lay Off 200-250 Employees In India Amid Global Restructuring

  Uber has announced a global restructuring that will cut about 3,300 roles worldwide and 200-250 jobs in India. - The company said the changes will remove management layers, consolidate functions, and affect technology and operations teams in India, with nearly half of the AI Solutions team reportedly impacted.

  • Uber also said it will ask most remote employees to return to the office, keep a hybrid policy of three office days per week, and invest more than $10 Bn in autonomous vehicle technology over the coming years.
• Sep 2, 2026

  🤖 La Machine #90: AI Regulation Gets Real

  The article is a newsletter-style roundup with one main first-time feature on INCOM and multiple separate headlines and event listings.

  • INCOM is highlighted as a French startup proposing a cooperative model for platform drivers under France’s “entrepreneur-salarié” system, framed as an alternative as France prepares to transpose the EU Platform Work Directive.
  • The rest of the piece briefly references other news items, including Nvidia/Hugging Face, EU AI enforcement, OpenAI, and the €387.8M Bull supercomputer contract in Finland, plus sponsored events in Paris.
  • Contacts are provided via email at the end of the newsletter: chris@frenchtechjournal.com and helen@frenchtechjournal.com.
• Sep 1, 2026

  Oracle To Lay Off Another 3,000 Employees In India: Report

  Oracle reportedly plans another round of layoffs in India. - Times of India reported that Oracle plans to cut nearly 3,000 jobs in India, with layoffs potentially starting September 1; Moneycontrol said the cuts could begin mid-September.

  • The report says this could eliminate up to 10% of Oracle’s 30,000-strong India workforce; Oracle has also previously downsized its India staff by nearly 12,000 in March citing AI-led automation.
  • The article adds that Oracle spent $55.7 Bn on capex in FY26, up from $21.2 Bn in FY25, to build data centres and server capacity for AI clients such as OpenAI and NVIDIA.