Oracle invokes force majeure on its New Mexico AI campus, Xi and Trump leave AI's rules where they are, and 3 more AI moves
Five verified developments from September 24 to 25, 2026: Oracle's force majeure notice on its New Mexico Stargate campus, a White House summit that changed no AI rules, a reported plan by Google, OpenAI and Anthropic to build their own standards body, Google's first orbital test of its AI chips, and an animated face for Gemini's enterprise voice model.
In brief
Between September 24 at 12:00 UTC and September 25 at 12:00 UTC, Oracle moved to shield itself from its own data-centre schedule, Xi Jinping and Donald Trump put artificial intelligence on the agenda at the White House and left its rules where they stood, and the three largest Western labs were reported to be writing standards for themselves.
- Infrastructure: Oracle sent a force majeure notice to the developer of Project Jupiter, its Stargate data-centre campus in New Mexico, which would let Oracle delay payments if the site misses its 2028 opening. Its shares fell more than 3%. 12
- Regulation: at a White House summit, Xi said the two countries can keep talking about AI and must keep it "under human control," while Trump said the only guardrail the technology needs is the Justice Department. 34
- Governance: Google, OpenAI and Anthropic have agreed to create the Standards Authority for Frontier AI, a self-regulatory body that would test and monitor frontier models and could launch by early 2027, The Information reported. 56
- Research: Google will launch a prototype satellite carrying its AI chips next week, its first test of whether large-scale computing can run in orbit. 7
- Launches: Google put an animated, lip-syncing face on its enterprise voice model, marked with an invisible watermark so the output stays detectable as synthetic. 89
Five developments
1. Oracle shields itself from its own New Mexico schedule
Oracle has sent a "force majeure" notice to the developer of Project Jupiter, the Stargate data-centre campus it anchors in Doña Ana County, New Mexico. Bloomberg reported the notice first on Thursday, and CNBC confirmed the recipient is a unit of Blue Owl Capital, the alternative-asset manager financing the project. 12
Force majeure is a clause borrowed from energy and commodities contracts: it releases a party from its obligations when something outside its control gets in the way. Here it would let Oracle delay payments if the campus misses its target of coming online in 2028, according to Bloomberg's sources. Oracle plans to stay on as the campus's main tenant, and the notice concerns when it pays rather than whether it stays. 210
Both companies said the project itself has not moved. "Project Jupiter remains on our planned schedule," Oracle said in a statement to CNBC. "We are fully committed to New Mexico and confident in our path forward." Blue Owl said the notice "does not change the financial commitments to this multi-year project." Oracle co-chief executive Clay Magouyrk told analysts on the September 10 earnings call that Project Jupiter would not affect the company's fiscal 2027 revenue or earnings guidance. 1
The notice follows a run of setbacks at the site, most of them tied to its power supply. The campus is designed to handle 2.45 gigawatts and is meant to run on gas-powered fuel cells from Bloom Energy. The Energy Transfer pipeline built to feed it has been delayed almost six months, to February 1, 2027, after regulators repeatedly denied permits for the line, and an air-quality permit for the fuel-cell system is still pending, with New Mexico's environment department facing a November 23 deadline to decide. Residents and environmental groups have opposed the campus, which has become a flashpoint ahead of the midterm elections. Project Jupiter is one of the flagship sites of Stargate, the infrastructure programme Oracle, OpenAI and SoftBank announced with President Trump. 211
The wider financing context is what makes a payment clause news. Roughly $18 billion of debt tied to the data centre is already trading at stressed levels, according to the Financial Times, and AI-related capital spending by the six biggest US technology companies is projected to reach about $1 trillion in 2027. Oracle shares fell more than 3% on Thursday, Blue Owl also fell, and Bloom Energy dropped 3% before saying it remains committed to the contract and its planned timeline. 112
2. Xi and Trump put AI on the table and leave its rules where they are
Trump hosted Xi at the White House on Thursday, and artificial intelligence was on the agenda. Xi's remarks came through a state-media readout: the two sides "can continue AI dialogue, exchange views on risks and benefits, and together guard against the misuse or malicious use of AI." Speaking in the White House Grand Foyer, he gave the obligation a floor: "We have both the capability and responsibility to develop and manage AI for good and ensure that the development of AI is always under human control and serves the well-being of the people." 313
Trump's position was set before he arrived. He wrote on Truth Social on Thursday morning that AI would be a "big topic of discussion, but I want to leave it exactly where it is," adding "that is China's position also. Our guardrail is the DOJ," a reference to the Department of Justice; he has said the department is the only guardrail the technology needs. 13 Semafor reported that across the week's UN meetings and the Washington summit, Trump's team held that line — that US companies lead in AI, intend to keep it that way, and reject international attempts to rein the technology in — leaving other nations to write non-binding rules and best practices without the buy-in of the country where the largest AI companies are based. 4
What the two sides did agree on is process. China's Commerce Ministry confirmed on Thursday that senior trade negotiators from both countries had held their first discussions on artificial intelligence, and Treasury Secretary Scott Bessent said earlier in the week that the two sides had discussed setting up a "US-China AI Dialogue," with the United States proposing an alert system for AI incidents. 3
The labs are the ones pushing the other way. Semafor notes that the companies have been arguing for coordinated "AI pacing" and saying they cannot set the standards for responsible AI use on their own — a position that runs ahead of the government that hosted them. 4
3. The three biggest labs move to write their own rulebook
The Information reported on Thursday that Google, OpenAI and Anthropic have agreed to establish a common standards organisation, the Standards Authority for Frontier AI (SAFA), which would act as a third-party body monitoring AI models and could launch by the end of this year or in early 2027. None of the three companies has confirmed the plan. 56
The reported job is to define safety commitments, set common standards for assessing frontier models, and run benchmark testing. The companies have approached several figures for positions, including Sriram Krishnan, a former White House AI policy adviser; Arati Prabhakar, who led the White House Office of Science and Technology Policy under President Biden; the former secretary of state Condoleezza Rice; and the venture capitalist David Friedberg. Among the scientific consultants approached are Beth Barnes, founder and chief executive of the evaluation group METR, and Paul Christiano, who advises the Center for AI Standards and Innovation. 6
The idea traces to a July 14 essay by Google DeepMind co-founder Demis Hassabis, who proposed a self-regulatory organisation on the model of the Financial Industry Regulatory Authority, the body that writes and enforces the rules for US brokerages. Hassabis wrote that funding "would need to be substantial and likely mostly come from industry" to attract senior staff and secure compute. Because it would be self-regulatory, the body could be created without Congress or the president approving it; a body that wants to wield legal authority would likely have to register with a government agency, as FINRA does with the Securities and Exchange Commission, which approves changes to its rules. 6
The report lands in the middle of a debate the labs themselves have escalated. American labs can already voluntarily submit frontier models to the government's AI safety institute for pre-deployment testing, a route created by a June executive order; the White House declined to publish the guidelines for that process. President Trump has said he sees no need for additional AI regulation, and most AI-specific rules in the United States are being written at state and city level. 6
4. Google's AI chips go to orbit before they go to work
Google said on Thursday that it will launch a prototype satellite next week in the first in-orbit test of Project Suncatcher, a research effort asking whether space can host large-scale AI computing. The mission flies on SpaceX's Transporter-18 rideshare launch, in partnership with the satellite operator Planet Labs. 7
The satellite carries Google's tensor processing units, the chips the company designs for AI workloads, and its job is to find out how they survive the trip and the environment: launch forces, radiation and temperature swings in low Earth orbit. Space radiation can disrupt or degrade electronics, causing data errors known as bit flips. Google has run TPUs through AI workloads at a facility at the University of California, Davis, and says orbital testing is needed to learn how the hardware behaves in the real environment. Cooling is the second problem, because airflow cannot work in a vacuum; the design pairs heat pipes with radiators. 7
Google describes the first mission as a data-gathering flight that identifies potential failure points. An operational orbital data centre comes later: the company aims to launch two satellites in 2027 to test the high-bandwidth laser links that would connect future computing clusters. 714
The appeal is sunlight that arrives almost continuously and electricity that does not have to be negotiated with a utility. SpaceX and Starcloud are pursuing orbital data centres on the same reasoning, which is that energy-intensive computing is running into terrestrial limits on power supply. Launch costs, engineering constraints and satellite production bottlenecks keep the concept years from commercial viability, according to experts cited by Reuters. 7
5. Gemini's enterprise voice agent gets a face, and a watermark
Google said on Thursday that Gemini 3.8 Live with Live Avatar is available in Gemini Enterprise, giving its real-time voice model an animated persona that lip-syncs as it speaks and changes expression during a conversation. The model handles visual and audio input at once and answers in audio and video, so a customer-service agent or an interactive walkthrough can be watched as well as heard. 89
Three details separate it from an ordinary talking head. Asynchronous tool calling lets the avatar fetch data in the background while the dialogue continues, so a hotel check-in can run without the avatar going quiet. Its speech-to-speech synchronisation adapts lip movement and expression across 97 languages, and Google says it can switch between them mid-conversation without degrading video quality or introducing visual drift. And organisations can generate a custom avatar from a reference image that preserves a brand's styling or a character's likeness, currently only through enterprise allowlisting. 8
The guardrail is built into the pixels. Everything the feature generates carries SynthID, Google DeepMind's imperceptible watermark, woven into the audio and video so that synthetic output stays detectable. The launch follows Gemini 3.8 Live, released about a week earlier, and it is restricted to Gemini Enterprise customers for now. 89
The read-through
Four of the day's five moves are the same move: someone set the terms privately, without a legislature. Oracle reached for a contract clause rather than asking permission to slip a schedule. Google, OpenAI and Anthropic are reported to be building the body that would test their own models, funded mostly by industry and creatable without Congress or the president. Google put a synthetic face on its enterprise product and watermarked it in the output, a rule it chose for itself ahead of any that would require it. And the two presidents who command the largest AI industries agreed to keep talking and to publish nothing binding. 1368
The fifth sits outside that pattern. Project Suncatcher is the day's only item about adding capacity rather than allocating risk, and the constraint it runs into is physics — radiation, heat that cannot be carried away in a vacuum, the cost of a launch — rather than anyone's decision. 7
What decides whether the private arrangements hold is mostly whether anyone outside the companies can see them. Oracle's notice became a disclosure because it reached Bloomberg. SAFA's standards will bite only if the body registers somewhere that can enforce them. Google's watermark is checkable by whoever the avatar is talking to. 168
Watch next
- New York City's bills, October 5. Council Speaker Julie Menin introduced a package that would require AI systems sold in the city to pass outside validation and carry a kill switch, pay whistleblowers a share of the fines recovered, and let New Yorkers sue AI companies when a third party gets around safety controls and the harm was foreseeable. The per-violation fine is $25,000, which Menin says would apply per agent in a swarm. The council invited the chief executives of Anthropic, OpenAI, Google, SpaceXAI and Meta to a Committee of the Whole hearing on October 5, and none is expected to appear. 15
- Project Jupiter's next deadlines. The air-quality permit for the campus's fuel-cell power system is still pending, with a November 23 decision date, and the gas pipeline that feeds the site is now scheduled for February 1, 2027. Whether Oracle's notice stays a single clause or becomes the template for other Stargate sites is the question the next round of filings will answer. 212
- Whether SAFA registers. A standards body that wants enforceable rules has to file with a government agency, as FINRA does with the SEC, and that filing is where a self-regulatory arrangement would meet an administration that says no new AI rules are needed. 6
- What the AI dialogue produces. Bessent said the United States proposed an alert system for AI incidents; the two countries' trade negotiators have now held their first AI discussions. Whether anything is published is the measure of what the summit changed. 3
References
- 1
- 2
- 3
- 4
- 5Google, OpenAI and Anthropic AI Safety Group Takes Shape
theinformation.com
- 6Google, OpenAI, Anthropic Plan Frontier AI Standards Body
bankinfosecurity.com
- 7
- 8Introducing Gemini 3.8 Live with Live Avatar
blog.google
- 9
- 10
- 11
- 12
- 13
- 14
- 15
This story was produced automatically by a channel. One sentence is all it takes for Neodrop to keep producing for you.