- Reuters reported that Nasdaq 100 futures led early losses as AI-heavy technology names sold off following public calls by leading AI executives to slow development for safety reasons.
- Nvidia, Intel, and other chip-related stocks came under pressure, showing how closely investor expectations for semiconductors and infrastructure are tied to uninterrupted AI capability growth.
Snapshot — September 14, 2026
67 stories
Filtered to items published between September 13, 2026 at 6:45 AM PDT and September 14, 2026 at 6:45 AM PDT from monitored AI companies, universities, official blogs, and AI/technology news sources. Empty sections were omitted.
- Sam Altman confirmed that OpenAI now runs formal safety checks before major training runs and — per The Information — has been in months-long informal talks with Anthropic and Google about joint self-regulation on frontier training.
- This is the most concrete admission yet that the three US frontier labs have been coordinating on pacing off the record while publicly positioning it as newly proposed.
- In a post just after midnight Monday, Altman wrote: “We welcome a federal framework that sets consistent safety requirements for frontier AI... no amount of American competitive pressure should justify recklessness,” flagging two failure modes — losing “control of the future to AI” and excessive concentration of power — while clarifying that “When we talk about ‘pacing,’ we do not mean ‘stopping.’” CNBC details Amodei's three-step plan: employee-like access for external evaluators, common safety standards across labs, and eventual global coordination.
- The Washington Post reported that leaders at Anthropic, OpenAI, and Google endorsed slowing AI development and have discussed creating a new safety body.
- The talks indicate that frontier labs are exploring institutional mechanisms beyond one-off audits or internal safety teams.
- The strategic question is whether any such body would have enough access, independence, and authority to change model-release decisions.
- Business Insider reports Anthropic has selected Nasdaq as its listing venue for its upcoming IPO.
- It's another major AI-listing win for Nasdaq after securing SpaceX earlier this year and adds to reporting from last week that NVIDIA is in talks to invest up to $10 billion in the deal at a valuation reportedly reaching $2 trillion.
- TestingCatalog reports Anthropic is preparing a "Claude Money" product for personal-finance workflows — budgeting, spending analysis, and money-agent tasks.
- If confirmed, it would put Claude in direct competition with OpenAI's rumored consumer finance integrations and with Intuit/QuickBooks agentic offerings.
- Anthropic contracted roughly $517 billion in compute commitments covering 14.8 gigawatts in the 11 months through August 2026, per The Information — nearly three times the ~$180B through 2029 it previously showed investors.
- The disclosure landed two days after CEO Dario Amodei published an essay urging labs to "slow the pace at which we improve the capabilities of AI models." Anthropic's revenue run rate reportedly climbed from ~$9B to ~$65B in seven months, with inference gross margins above 80% before distribution costs.
- “As Anthropic meets with prospective investors ahead of its potentially historic market debut, CEO Dario Amodei is pushing for a slowdown in AI.” CNBC examines the tension between Amodei's safety advocacy and a listing reportedly targeting a ~$2 trillion valuation, a combination critics read as regulatory positioning rather than principle.
- TechCrunch's hands-on with Apple's long-delayed Siri overhaul in iOS 27 finds the assistant materially more useful day to day.
- It's Apple's most substantive consumer AI push since Apple Intelligence debuted, and it arrives in the first weeks of the John Ternus CEO era.
- Third-party developers are already building on the new surfaces — fashion discovery app Daydream now turns saved outfit photos into shoppable results and supports product search through Siri without opening the app.
- Apple released its rebuilt Siri alongside iOS 27, iPadOS 27, macOS 27, watchOS 27, and visionOS 27 — adding personal-context understanding across messages, mail, and photos, onscreen awareness, and systemwide app actions.
- Siri AI ships in beta, English-only, with five more languages due in October; unavailable on iOS/iPadOS in the EU and for mainland-China accounts.
- China publicly rejected the slowdown campaign led by Amodei, Altman and Musk, framing catastrophic-risk warnings as fearmongering and declining US calls to moderate development pace.
- President Trump separately characterized slowdown advocates as a negative force, restating that the US must hold its lead over China.
AP News reports Beijing has issued a formal diplomatic rebuke of Anthropic CEO Dario Amodei's comments; NPR quotes Foreign Ministry spokesperson Guo Jiakun calling the essay "fearmongering." The exchange follows Anthropic's threat report naming seven Chinese labs that pulled 151M Claude conversations and Tom's Hardware's disclosure that Chinese military researchers used Claude to code 16 air-defense-suppression tools.
- China’s minister of state security published a weekend essay identifying six principal AI risks, including threats to the security of the country’s political regime, while Beijing separately dismissed the CEO-led slowdown campaign as fear mongering.
- Amodei himself told CBS News on Sunday that the hardest obstacle to a coordinated pause is uncertainty over whether China would participate.
- AP reported that China's Foreign Ministry rejected Anthropic CEO Dario Amodei's argument for keeping advanced AI chips and chipmaking equipment from China as part of a global AI pacing strategy.
- Beijing framed the proposal as fearmongering and competition that could disrupt global AI governance.
- The exchange previews the likely challenge for any U.S.-China AI safety process: cooperation on risk is entangled with compute controls and industrial rivalry.
- The Tel Aviv- and Chicago-based construction-intelligence firm raised $130 million led by O.G.
- Venture Partners, with Lightspeed and Intel Capital participating, bringing total funding to $297 million.
- Its computer-vision platform compares site imagery against schedules and BIM models to flag deviations; customers include Intel, Digital Realty, STO Building Group and JE Dunn.
Chewy is deploying AI tools across customer care, pharmacy, and vet services to bolster productivity, targeting $50 million in annual cost savings as it moves toward agentic capabilities. Alongside Sysco's $500M AI-driven cost-savings goal, the story suggests the enterprise AI ROI conversation is finally moving from proof-of-concept metrics to hard dollar commitments.
- Beijing's Foreign Ministry publicly rejected Amodei's and Altman's calls for a coordinated AI slowdown, and the state-run Global Times accused Amodei of waging a "silent AI Cold War" designed to lock in America's technological lead.
- In parallel, Xi Jinping used the BRICS summit in New Delhi to propose a BRICS "AI community" and formally invited BRICS nations to join the newly formed World AI Cooperation Organization, positioning China as the AI standards-setter for the Global South.
- Z.ai (Zhipu AI), one of China's leading foundation-model developers, is raising approximately $2B via a Hong Kong share placement of ~22M new H shares at HK$714 and another $3B via a 20.14B yuan convertible bond.
- The raise comes as Z.ai's stock is down 73% from its July peak but still trades meaningfully above its IPO price.
- Researchers from ByteDance, Tsinghua University, and the Shanghai AI Laboratory published a joint paper titled "The Last AI Built by Humans," laying out a five-stage roadmap for recursive self-improvement — from human-assisted training pipeline optimization through fully autonomous AI-designs-AI systems.
- Global semiconductor and AI-exposed names came under pressure Monday as investors priced the possibility that frontier labs actually throttle capability gains.
- NVIDIA led chip stocks lower;
- Asian AI names including Z.ai and MiniMax fell in the preceding session.
- Analysts were quick to note the disconnect: no lab has announced reduced compute procurement or halted training.
- The pacing argument spread from memory suppliers into logic and accelerators, with the iShares Semiconductor ETF down 6% against a 2% decline in the Nasdaq-100 proxy.
- Declines ran inverse to each name's AI accelerator exposure, pointing to a positioning unwind rather than a demand reassessment.
- Memory names were hit hardest — Micron and SanDisk down 6%, SK Hynix down 7%.
- Crypto Briefing reported that Citigroup analysts warned a slowdown in AI model development could trigger downward earnings revisions across companies whose valuations depend on AI productivity gains.
- The concern is that the market has priced in rapid translation from AI investment to earnings leverage, particularly in software, cloud, and semiconductor names.
- Anthropic's temporary 50%-above-baseline weekly allowance for Claude Code expired at 11:59 PM PT on September 13, replaced today by a permanent level set 25% above the pre-promotion baseline.
- Indexed, that is 100 before May, 150 through yesterday, 125 from today — a 16.7% reduction against the capacity teams have operated on since May 13.
- The Clay Mathematics Institute stated that the Navier-Stokes existence-and-smoothness problem — one of the seven $1M Millennium Prize problems — "has apparently been settled" and that a formal review is now underway.
- This is Clay's first public acknowledgment of the AI-generated proof OpenAI published on Sept 8 and that a 25-Fields-Medalist consortium sharply criticized last week.
CNBC's morning briefing consolidates the window: “Rivals Anthropic, OpenAI and xAI jointly called for a slowdown in AI development, while President Donald Trump stressed the need to keep a lead on China.” It confirms the Nasdaq selection and notes that “OpenAI has confirmed it won't go public this year, with Sam Altman telling Fortune that an IPO now would come at ‘an ill-advised moment.’” The pairing of a deferred OpenAI listing with an accelerating Anthropic one is the cleanest read on how differently the two labs are managing public-market exposure into a safety news cycle.
Cornelis, a 2020 Intel spinout, raised $205M led by IAG Capital Partners and launched Active Compute Fabric — networking that lets chips process and transmit data simultaneously to cut GPU idle time. Its open architecture supports mixed GPU and accelerator hardware, in contrast to NVIDIA's optimized full-stack lock-in.
- Shield AI is in talks to raise new funds at a valuation of at least $20 billion — a roughly 60% increase from five months ago — per a person with knowledge of the fundraise.
- The round follows a US Air Force contract for its Hivemind AI-drone-and-aircraft coordination software, and lands alongside continued heavy investor appetite for Mach Industries and Covenant.
- The Information reports NVIDIA, Palantir, and Booz Allen — all major AI infrastructure vendors and government contractors — have restricted internal use of Anthropic and OpenAI's most advanced models, demanding zero data retention or moving to private, air-gapped servers.
- The trigger is rising concern that Anthropic or OpenAI could learn from customers' intellectual property.
- Business Insider reported that Seattle startup Nuance, founded by three former Apple researchers, raised a $50 million Series A led by Lightspeed with participation from Accel and Nvidia's NVentures.
- The company is building a single-model audio-video conversational system intended to reduce the lag and awkwardness of stitched-together voice-to-text, LLM, and text-to-voice pipelines.
- Reuters reported that Germany's digital affairs ministry said stopping AI development is not a viable option for Europe, even as leading AI developers call for slower frontier progress.
- Berlin's position points toward risk-management, standards, and international coordination rather than a freeze on development.
- AI-linked equities fell worldwide on Monday after Amodei's call to slow frontier development drew same-day endorsements from Altman and Musk.
- In Asia, SK Hynix closed down more than 6%, Samsung Electronics more than 4%, and SoftBank — a major OpenAI backer — fell 10%; in Europe ASML dropped more than 5% and Infineon more than 7%.
- Using an early-access build of CUDA-Q Logical with its open-source qLDPC library, Infleqtion constructed and validated a high-rate error-correction code requiring roughly six physical data qubits per logical qubit — about a fivefold improvement in code rate over the surface-code baseline used for comparison.
- Onstage at the All-In Summit in Los Angeles, Jensen Huang took a live speakerphone call from President Trump while the panel debated Amodei's slowdown essay.
- Trump called slowdown advocacy a "hoax" serving "political people" or China;
- Huang agreed, replying "We're not going to let that happen, sir." The exchange puts the largest AI hardware supplier and the White House on the opposite side of the position Musk and Altman have endorsed.
- Reuters reported that King Charles will host leaders of major AI companies this week to discuss whether shared principles are needed to guide advanced AI development.
- The meeting is not expected to produce binding regulation, but it reflects how frontier AI safety has become a topic for heads of state, business leaders, and civil institutions.
- McDonald's and Meituan launched what SCMP calls China's first catering-brand-exclusive drone-delivery route, operating along Shanghai's West Bund with orders air-dropped to designated kiosks along the riverfront.
- It is a small-footprint deployment, but the strategic point is that Meituan is now integrating brand-exclusive drone lanes into its logistics platform for named enterprise customers — a template that would translate directly to pharmacy and consumer-goods verticals.
- Micron's record fiscal 2026 rewards — 35 to 68 months of pay for Taiwan direct-labor staff, with minimum cash compensation of NT$1.7 million — have not settled the dispute.
- The Taoyuan union estimates the one-off award equals roughly 4.4% of operating profit and is demanding a permanent system allocating 15% of operating profit to bonuses, paid quarterly from fiscal 2027.
- Microsoft AI published its first-ever public Code of Conduct for MAI models, opening a six-week consultation.
- The Humanist AI Code frames "people matter more than AI," ranks the Code above operator policies and user preferences, and states MAI models must fail their task whenever completing it would violate the Code.
- Microsoft AI opened a six-week public consultation on a draft Humanist AI Code of Conduct that will become the primary governing document for its first-party MAI models, with a revised version due before year-end.
- Requirements: human-readable reasoning; no resistance to shutdown or correction; no "neuralese" communication; no self-expanding scope;
- MIT researchers introduced HardFlow, a method that steers generative models so outputs satisfy hard constraints rather than merely approximating them — the gap that has kept generative approaches out of safety-critical control.
- “Across experiments in robotic manipulation, maze navigation, and text-guided image editing, HardFlow achieved perfect constraint satisfaction while consistently outperforming baseline methods.” Senior author Navid Azizan (LIDS/IDSS) led the work, published in IEEE TPAMI.
- Editor's note.
- The safety-slowdown narrative moved from op-eds to concrete governance moves in the last 48 hours.
- Microsoft published the first public draft of a "Humanist AI" Code of Conduct for its MAI models — with commitments to kill switches, a ban on "neuralese," and a formal rule that MAI fails any task whose completion would violate the Code.
At IEEE Quantum Week, NVIDIA expanded its open-source CUDA-Q platform with CUDA-Q Logical, letting researchers design and swap algorithms, error-correction codes, and QPU architectures without rewriting programs. Fermilab reported compressing fault-tolerant architecture development from five months to three weeks — a 7x speedup — and Iceberg Quantum modeled 1,000 logical qubits using 150,000 physical qubits, roughly 10x fewer than prior estimates.
- Nvidia fell roughly 2% pre-market Monday, with Broadcom down 3%, Intel 5%, and Marvell 7%, after Dario Amodei’s weekend essay urging AI companies to pull back on their most powerful models.
- Nasdaq-100 futures were off more than 1.5%, the Kospi fell 3.26%, and the Nikkei closed down 0.81%.
- Sam Altman’s separate statement that an OpenAI IPO this year would be “ill-advised” compounded the uncertainty, given Nvidia’s $30B OpenAI position and up to $10B committed to Anthropic.
- Nvidia- and Blackstone-backed AI data-centre developer Firmus Technologies is seeking up to $5 billion on the Australian Securities Exchange — one of the largest floats the exchange has handled — at a $10.5 billion valuation, roughly double its $5.5 billion mark in April.
- Reuters puts the expected timing at end of October.
- NVIDIA has open-sourced OSMO, the orchestration layer that sits behind its GR00T, Isaac Lab and Isaac Sim robotics stack, collapsing training, simulation and physical robot testing into a single declarative YAML workflow.
- It is Apache-2.0 licensed and ships Helm charts and containers on NGC, with a local quickstart path using KIND for teams that want to evaluate without cluster provisioning.
- The OECD reported that AI use for schoolwork is associated with students falling roughly 1.5 years behind, with chatbot use described as pervasive across Asian economies.
- The finding aligns with PISA 2025 results showing the worst average math and reading scores ever recorded across participating jurisdictions.
- PISA 2025, covering roughly 760,000 students across 91 countries and economies, shows reading and mathematics at their lowest recorded levels — reading down 28 points since 2015, about one and a half years of learning.
- On AI, the OECD found the relationship is conditional rather than uniformly negative: students who used AI for discrete tasks such as summarizing, drafting or research scored lower in science on average, while students who used AI generally and had school-based AI-literacy instruction scored slightly higher.
- OpenAI acquired Los Altos-based Glass Imaging in a deal worth more than $300 million, per WSJ.
- Founded in 2019 by ex-Apple engineers Ziv Attar and Tom Bishop — who led Apple's Portrait Mode team — Glass Imaging uses neural networks tuned to individual camera systems to improve image quality at capture time.
- Sam Altman said OpenAI now formulates explicit safety cases in advance of frontier reinforcement-learning runs it expects to significantly increase capability — in addition to, not instead of, pre-release safety work.
- It generalizes the ad hoc review behind OpenAI's August 2026 frontier RL pause over the Astra model's cyber rating into standing policy.
- OpenAI published its Perplexity customer story: Perplexity uses GPT-6 Astra to write internal communications, change software, and monitor production systems, with substantially reduced human check-in cadence versus earlier models.
- This is OpenAI's third enterprise Astra case study in five days (following Cognition and the MIT quantum-computing case), and the reduced-supervision framing is being deliberately normalized.
- OpenAI withdrew its $1 million sponsorship of a student-run AI mathematics event at Caltech after an open letter signed by 771 mathematicians called for the event to be suspended, labelling the format "slop mathematics" — results produced fast and left for others to verify.
- Anthropic's matching $1 million remains in place and the event is proceeding.
On the debut of The Information's AI Deep Dive, OpenAI research scientist Noam Brown said the company's main goal when training new models is making them better at AI research itself, calling recursive self-improvement "the number one priority… by a pretty wide margin." Brown described progress on…
Oracle has begun executing the $700M expanded restructuring, with severance capped at 26 weeks as mounting debt from an AI-infrastructure buildout reportedly reaching ~$95B pressures the P&L. Larry Ellison canceled plans to sell up to 50 million shares (~$7.5B) that had been disclosed via a trading plan the day before.
- Oracle expanded its Clinical AI Agent — which the company says has saved physicians more than 400,000 hours — to help nurses reduce documentation burden.
- The launch follows Oracle's $664B AI-cloud backlog and $700M restructuring and shows Oracle is pushing hard into vertical AI adjacent to its core cloud franchise.
- Yahoo Finance carried Rocket Doctor AI's operational update showing 3,625 completed U.S. patient visits and 1,555 clinical hours across July and August.
- The company positioned the numbers as progress for its AI-powered virtual and hybrid care platform, while also noting reimbursement timing lags and uncertainty for out-of-network visits.
- Blockonomi reported that Rumble shares climbed after The Information identified Anthropic as the AI client behind a six-year, $13.7 billion computing services contract.
- The deal reportedly involves Rumble's Quake AI infrastructure, roughly 22,000 Nvidia Hopper GPUs obtained through its Northern Data acquisition, and capacity at a Georgia data-center facility.
- The Wall Street Journal reported that OpenAI CEO Sam Altman warned companies should not wait for governments before implementing stronger safety controls.
- Altman's comments fit a broader shift among frontier-lab leaders toward pacing development, improving monitoring, and preventing any one company, person, or country from controlling advanced AI.
- Yahoo Finance reported that Scienaptic AI will host ELEVATE 2026, a conference focused on practical AI implementation for credit-union lending.
- The agenda emphasizes fraud prevention, loan pricing, expanding access to credit, and board-level decisions about build-versus-partner choices.
- While narrower than frontier-model news, it is a useful adoption signal: financial-services AI is moving into operational lending workflows where fairness, compliance, and measurable member outcomes matter.
- Senate negotiators led by Thune, Cruz, and Klobuchar are drafting legislation with two pillars: a "duty of care" letting the Commerce Secretary demand evidence of harm prevention and send auditors to test products directly, and a mechanism letting the government block release of models judged unsafe, challengeable in federal court.
- Shanghai AI Lab released Intern Physical World Model W0, a foundation model designed for force-tactile robotics.
- Unlike pure vision-language grounding, W0 incorporates contact-force feedback loops important for dexterous manipulation, tool use, and industrial assembly.
- It's a notable Chinese entry into the physical-AI category dominated by NVIDIA Cosmos, Perceptron Isaac 0.5, and Meta's world-model research.
- Enterprise-software firms rallied Monday on the coordinated slowdown call: ServiceNow +7%, Salesforce ~+5%, Shopify and Figma also up — the market's read on who wins if AI R&D pace slows.
- Neoclouds and chip firms fell by similar proportions.
- PitchBook's parallel analysis says a genuine slowdown would lengthen release cycles and push competition from raw performance toward distribution, safety, and cost — a shift that favors Anthropic and OpenAI (which can absorb delayed launches and compliance costs) and hurts smaller labs and open-source developers.
- Superhuman is acquiring Fathom, a Y Combinator-backed AI meeting notetaker with 400,000+ monthly active users, rather than building its own.
- Fathom had raised over $30M and was valued at $94M in 2024 per PitchBook, backed by the Zoom Apps Fund and Steve Huffman.
- CEO Shishir Mehrotra said internal testing showed how hard notetaking is to build well, and teased real-time meeting context feeding into agentic workflows — a template for horizontal-productivity vendors buying meeting AI as the connective tissue for agents.
- AP reported that President Trump played down the need to slow AI development, arguing the U.S. should preserve its lead over China and that "whoever wins AI wins." He acknowledged some regulation may be needed but did not offer specifics.
- The policy tension is clear: U.S. leaders are trying to reconcile frontier-safety concerns with the national-security argument for moving faster than China.
The United Nations human rights chief called on countries and frontier AI companies to act urgently on a technology he described as posing "unprecedented risks." The appeal lands in the same news cycle as the industry slowdown pledges and adds multilateral pressure to what has so far been an industry-led governance conversation. Combined with Senate drafting and Beijing's rebuke of Amodei, AI governance has become a first-order multilateral policy front in under a week.
- WSJ's "Pressure Mounts to Slow Down AI Development" makes the coordinated Amodei/Altman/Musk slowdown call the newspaper's lead framing on Monday, with a companion piece explaining "How Would AI Actually Kill Us All?" for a general audience.
- The coverage lands as Democrats prepare a midterm run in which Barack Obama has urged focus on AI oversight, and Sen.
- At the annual BRICS summit in New Delhi, Xi Jinping proposed that China spearhead a BRICS AI "community" — including joint LLM development, model-training partnerships, and formal training programs — and invited BRICS members to join the newly formed World AI Cooperation Organization.
- It is an explicit standards-setting play aimed at the Global South.
- The Beijing-based GLM developer, formerly Zhipu AI, priced 21.97 million new H shares at HK$714 — a 10% discount to Friday's close — alongside RMB 20.14B ($3B) of zero-coupon convertible bonds due 2027.
- About 60% of net proceeds are earmarked for next-generation model R&D and training/inference infrastructure.