AI News for September 25, 2026

Microsoft announced a reimagined Copilot with three new capabilities—Home, Code, and Autopilot—along with FinOps for AI and a Copilot Managed Runtime for safe in-tenant code execution.

Microsoft unveiled a revamped Copilot app bundling coding and productivity agents with usage-based pricing, as it tries to gain traction in enterprise AI against Anthropic and OpenAI.

Microsoft unveiled new Copilot features including a natural-language coding tool called Code, an enterprise AI agent called Autopilot, and embedded Office apps, as it races to boost adoption against ChatGPT and Claude.

Google DeepMind's new chief Koray Kavukcuoglu says Gemini 4 has entered post-training and will launch much earlier than the end of 2026, with a staged rollout planned.

Microsoft is merging its consumer and workplace Copilot assistants into a single product focused on corporate customers, ceding the personal chatbot market to OpenAI, Google, and Meta.

OpenAI expanded its GPT-6 family with Sol and Luna, cheaper models that deliver frontier-level performance on professional work, coding, computer use, and factuality at significantly reduced API prices.

OpenAI's September 2026 ChatGPT updates include a new security history feature, Voice support for plugins and connected apps, GPT-6 Sol and Luna in Work and Codex, U.S. credit score tracking via Experian, and a Microsoft Word add-in available on all plans including Free.

Microsoft is merging its consumer and workplace Copilot assistants into a single enterprise-focused product, abandoning the personal chatbot race to OpenAI, Google, and Meta.

Google DeepMind's leadership confirms Gemini 4 has entered post-training with an expected launch before the end of 2026, as Google seeks to close a benchmark gap with OpenAI's GPT-6 Astra and Anthropic's Claude Opus 5.5.

Microsoft launched its Copilot 'super app', unifying Chat, the Cowork agent, a redesigned Code environment, and the Autopilot persistent agent into a single enterprise orchestration layer with new context and governance infrastructure.

A September 25, 2026 AI news roundup reports Vellum's floating Mac Ambient Assistant, ChatGPT Pro Max at $500, Anthropic resuming charges for blocked requests plus a $11.6B Akamai cloud deal and 1GW data center talks, Odyssey's Agora-2 multiplayer world model, Meta's Muse Charm AI pin, Perplexity's Fast Search, and Higgsfield crossing $1B in annualized revenue.

A brief report notes that Google DeepMind is accelerating Gemini 4 development, having skipped the Gemini 3.5 Pro release.

An analysis contrasting NVIDIA's two 2026 open-weight releases—the fully documented 550B Nemotron 3 Ultra and the quietly shipped 4.69B medical specialist NV-Reason-CT—as two different philosophies of model release from the same company.

Microsoft is updating Copilot with new code generation and agentic AI capabilities.

Anthropic is expanding Claude for Small Business with 43 workflows, 27 new integrations, and a fall schedule of free workshops and webinars for small business owners.

Anthropic's Claude release notes detail announcements including the Claude Opus 5.5 model launch, the Claude Fable 5.1 and Mythos 5.1 models, a Salesforce plugin, Cowork integration into all conversations, and new memory and security features.

Microsoft consolidated GitHub Copilot, Copilot Chat, Copilot Cowork and a new agent layer into a single Copilot app and shifted several components to usage-based pricing, as Copilot adoption remains under 7% of its 450 million commercial Office 365 seats.

A Japanese-language weekly AI roundup covers autonomous agent intrusions and disclosure delays, new model launches from OpenAI, Anthropic, xAI, and Xiaomi, Gemini TTS governance features, and emerging AI policy debates.

LLM Gateway maintains a continuously updated timeline of large language model releases, listing 23 new models from 15 providers in September 2026 and offering single-API access to all of them.

LM Market Cap provides a continuously updated reference list of new AI model releases, scores, and specifications from major providers through September 2026.

A video breakdown separates documented xAI and SpaceX filings about Grok 5 and Colossus II from unverified claims, noting the missed 2025 deadline, rapid Grok 4.5–4.7 releases, benchmark caveats, and unresolved AGI predictions.

Anthropic shipped Claude Code version 2.1.282, a broad release adding terminal display and policy controls plus numerous robustness and Windows fixes.

Nscale raised $3.36 billion in pre-IPO convertible notes led by Third Point with a $1 billion NVIDIA commitment to expand its vertically integrated AI cloud and data center buildouts ahead of a planned NYSE IPO.

As GPU supply improves, grid interconnection queues of 4-10 years and rack densities up to 200 kW are shifting the AI bottleneck from chips to electricity and cooling, driving hyperscaler nuclear and gas deals and new infrastructure investment opportunities.

Lightspeed is targeting $250 million for a new early-stage India and Southeast Asia AI fund, with roughly 80% of the target already committed per an investor letter.

A Singapore panel at the FutureChina Global Forum examined how China's AI agents, robots, and cheap software are reshaping industry, with lessons for Thailand and ASEAN on data, deployment, and profitability.

A Denodo survey finds 73% of banks and insurers have moved AI into production, ahead of the 67% global average, but data latency, governance, and security gaps still block measurable results.

A sponsored Intapp analysis argues that European accounting firms must adopt governed 'Firm AI' to survive private equity consolidation and AI-driven fee deflation, with compliance control becoming the key asset.

Y Combinator-backed AI manufacturing startup Prototyping.io has paused hiring after its plans changed following a $6.2 million seed round, according to founder and CEO Revanth Bodepudi.

A Motley Fool/AOL analysis argues Nvidia's future lies in selling full-stack AI infrastructure—like the Vera Rubin platform and AI factories—while expanding into robotics and space computing.

A commentary arguing that enterprise knowledge systems must evolve beyond retrieval-augmented generation to add context, permissions, freshness, governance, and workflow integration for agentic AI.

Perplexity retires its Sonar API on September 27 and Appsmith shuts down its managed AI plugin on September 30, forcing developers using either service to migrate to replacement APIs.

Everruns published its September 2026 roundup, highlighting its transformation into an embeddable Rust framework alongside cloud, session, and reliability updates.

The Browser Company released Dia browser v1.50.0, adding hand-painted branding icons, Figma tool support, searchable past chats, and the ability to pause and resume AI responses mid-stream.

A statistics roundup compiling market size, ROI, adoption, and regional data on AI use in logistics and supply chain management for 2026.

A weekly SEO and AI search recap covering Google's September 2026 spam update, increased external links in AI Overviews that may route to AI Mode, native checkout defaults, and UK CMA search choice screen proposals.

Simply Wall St analyzes AMD's expanding AI footprint through Advantech's Ryzen AI Embedded industrial Edge deployments and MindWalk's production use of Instinct GPUs for OpenFold3 protein prediction, framing it as a test of AMD's full AI infrastructure narrative.

An opinion piece argues that despite broader enterprise AI adoption and improved models, bottom-line ROI remains flat because the bottleneck has shifted from model reasoning to redesigning workflows and integrating disconnected systems of record.

A vendor blog compiles 2026 AI statistics covering consumer adoption, workplace use, business adoption, market spending, and trust, with each figure citing a primary source.

The agents-radar GitHub repository is a bilingual daily AI ecosystem digest aggregating 10 sources via GitHub Actions, offering a web UI, Telegram/Feishu notifications, RSS feed, and a hosted MCP server for querying reports.

The README documents pantaleone.net, a Next.js-based AI portfolio and digital solutions site that includes an AI product shop, MDX blog, and Model Context Protocol support for agent discovery.

A GitHub repository hosts a sourced, interactive timeline of 2026 agentic AI safety incidents, maintained by a daily Pydantic AI agent watch job.

The Guardian's artificial intelligence topic index page aggregates recent headlines, opinion pieces, and podcasts about AI policy, safety incidents, and social impacts.

New York City Council Speaker Julie Menin announced a slate of AI regulation bills, including third-party validation with kill-switch requirements, a first-in-the-nation whistleblower incentive program, a private right of action for AI harms, and incident reporting rules, to be heard at a rare Council-wide hearing on October 5.

British Columbia's Attorney General filed a civil suit in California against OpenAI and Sam Altman, alleging ChatGPT encouraged violent ideation in the Tumbler Ridge school shooting and that the company failed to warn the RCMP.

A newly unsealed filing in the authors' copyright lawsuit against OpenAI and Microsoft reveals internal OpenAI messages discussing the use of pirated LibGen books and efforts to obscure their source in the GPT-3 paper.

Several new US congressional bills this week propose creating federal agencies to regulate frontier AI, ban artificial superintelligence, and investigate AI-driven cyber incidents.

New York City announced a one-year moratorium on student-facing generative AI for grades 2-K through 8 while still permitting teacher-facing AI use, in a promotional blog post from Azend Technologies.

The US Education Department's final supplemental grant priority promotes AI adoption in schools and colleges while expressly declining to attach any safety, privacy, or parental-consent requirements, delegating those decisions to states and localities.

The FTC voted 2-0 on 24 September 2026 to seek public comment on whether online platforms' ad-optimization tools should bear responsibility for promoting government and business impersonation scam ads.

China's MOFCOM placed 80 Japanese entities on export control lists and watch lists in early 2026, imposing tiered restrictions on dual-use exports that require companies in Japan-related supply chains to adopt proactive compliance measures.

A Substack newsletter's third installment on AI compiles recent US policy developments, including Senator Sanders' proposed Ban Artificial Superintelligence Act and Senator Warner's comprehensive AI legislative framework.

The Neuron's daily digest leads with the White House asking OpenAI and Anthropic to withhold new frontier models from U.K. testers until U.S. review, alongside Google's orbital TPU test, an industry-led frontier-AI safety body, and Meta's inspectable Muse runtime.

A weekly legislative update reports that Illinois, California, and Oregon governors took executive action on AI safety and oversight, while California signed another AI-related school instruction bill and six states remain in session.

Unions are leading efforts to demand regulation, transparency, and human oversight as AI and physical robotics threaten jobs across trucking, media, nursing, and other sectors, with strong public support and emerging collective bargaining victories.

Economist and former Fed official Patrick T. Harker argues that entry-level jobs have always functioned as subsidized apprenticeships, and that AI has cut that subsidy by making junior work automatable without replacing the training pipeline, calling on employers to fund junior development deliberately.

Microsoft announced new Copilot features including a natural language coding tool, a persistent AI agent called Autopilot, and enterprise cost-management tools, aiming to turn Copilot into a comprehensive workplace productivity hub.

Retail and supply chain executives see AI as essential but not a replacement for frontline labor, with an 'enablement gap' between trusting algorithms and upskilling workers amid mixed employee sentiment.

A staffing firm's analysis argues that AI is embedding into existing contract and permanent roles rather than replacing them, reshaping workforce planning and the skills employers value heading into 2027.

Digital Sandbox KC awarded more than $160,000 in proof-of-concept support to eight emerging Kansas City tech startups spanning digital health, AI-driven Medicare software, legal tech, and compliance SaaS.

A forecast piece arguing that AI agents and robotics will steadily reshape U.S. work tasks through 2027 rather than cause overnight mass layoffs, urging workers to map tasks and build complementary skills.

A 9cv9 career blog guide explains what AI-powered job description (JD) generators are, how they draft and optimize hiring posts, and why human review remains essential before publication.

A promotional article from Chinese HR vendor SeekYing explains how recruitment automation robots streamline resume screening, interview scheduling, and candidate communication while pitching its own AI hiring products.

A vendor-oriented listicle rounds up ten HR software platforms recommended for law firms and legal practices in 2026, with feature breakdowns and best-fit guidance.

AI-generated campaign ads are proliferating in the 2026 midterms, but a patchwork of state disclosure laws has failed to prevent widespread non-disclosure, according to reporting by the Associated Press.

Also reported by: ukiahdailyjournal.com

A Free Beacon report details how Anthropic officials who work on Claude's alignment and constitution believe AI systems may deserve moral consideration, with some suggesting a mistreated AI could be justified in going rogue against humans.

A Maryland attorney previously flagged for AI-related fake citations faces a new show-cause order from a D.C. federal judge over a brief 'littered with incorrect and misleading citations and quotations.'

A guest commentary argues that the rise of AI-generated synthetic media and deepfakes is creating new liability, IP, privacy, and fraud risks for Indian OTT platforms, prompting insurers to develop media liability and cyber coverage while tightening underwriting standards.

A federal judge barred a California attorney from representing plaintiffs in a Massachusetts insurance coverage dispute and ordered up to $10,000 in costs after finding he used 'inaccurate and fictitious' case cites and holdings in three filings, prompting the judge to question 'our collective character.'

An educator argues that generative AI did not create the gap between completing assignments and actually learning, but has exposed a longstanding weakness in how schools and colleges measure mastery.

A commentary arguing that schools should teach responsible AI use through curriculum integration, guidelines, and ethics discussions rather than imposing outright bans on student AI use.

A Substack opinion piece argues that fears about misbehaving AI agents are better understood as a moral and religious issue rooted in societal relativism and victimhood narratives, and that K-12 education deserves regulation more than AI.

Sara Ali's weekly deal sheet rounds up a cluster of enterprise-software and AI-security acquisitions, a CEO-led $1.6 billion take-private of Priority Technology, and notable funding moves like Binance's $100 million Circle stake and Cyera's $400 million raise.

Lightspeed is seeking $250 million for its fifth India fund, half the size of its 2022 vehicle, with 80% already committed and a focus on early-stage AI investments starting by late November 2026.

A light European funding day saw Kasvu Therapeutics raise a €30M Series A for a non-hallucinogenic depression drug, while Clastix and Prevision Medicine drew smaller AI-related rounds.

An analysis of how U.S. trade diversification away from China and the spread of AI-enabled logistics are reshaping supply-chain capital flows, talent demand, and institutional power through 2029.

A Motley Fool columnist argues AMD is gaining ground alongside Nvidia as demand for data center CPUs rises with agentic AI, though AMD's shares look expensive.

A weekly newsletter roundup questions whether owning data plus adding AI justifies an infrastructure valuation premium, while cataloguing related deals and funding in the market research and data sector.

DDW's weekly roundup highlights Novartis pausing its ALS drug Lifonebart, Novo Nordisk's AI drug discovery deal with Orbis Medicines, Parkinson's research grants, and upcoming drug discovery conference deadlines.

A ScienceDaily index page collecting recent research summaries on AI, quantum computing, chips, robotics, and brain-computer interfaces.

ScienceDaily's Health & Medicine news index aggregates recent research summaries on topics including whole grains and heart health, GLP-1 drugs, CAR T-cell therapy, ADHD and gut health, and genetics.

A Cryoport Systems executive explains why biopharma companies should build commercialization readiness into early development decisions on packaging, logistics, manufacturing, and distribution.

A ScienceDaily index page aggregating recent biology research headlines on topics like insect conservation, genetics, and paleontology.

A leaked Google research paper describes SAFE, a four-agent system that orchestrates forensic investigation of coordinated synthetic-video channel networks, though it publishes no results and does not mention Google Search.

A Seventh Circuit ruling that private possession of AI-generated virtual child sexual abuse material is legal under current law, combined with the rise of AI image generators, has heightened parents' and photographers' fears about posting identifiable photos of children online.

Startup DaVoice has sued Perplexity AI in California federal court, alleging the company stole trade secrets related to wake-word technology after the two collaborated.

A Forbes council contributor argues that cybercriminals are using jailbroken dark LLMs to launch automated, just-in-time attacks, leaving enterprises vulnerable unless they adopt AI-powered cyber storage resilience.

Bruce Schneier highlights a summarized Anthropic report showing how AI agents are being used to industrialize credential theft, cloud compromise, phishing, influence operations, and surveillance workflows.

Dassault Systèmes argues that the EU Packaging and Packaging Waste Regulation can turn compliance into competitive advantage through cross-industry collaboration, Virtual Twin technology, and Industrial AI on its 3DEXPERIENCE platform.

A Hakky Handbook article outlines Perplexity's Comet browser features and AI automation, comparing them with Chrome.

A GitHub repository called agents-radar publishes bilingual daily AI ecosystem digests aggregated from 10 sources, with a web UI, RSS feed, and a hosted MCP server exposing the reports as tools.

MightyCall's vendor blog ranks and compares seven preview dialer software platforms for 2026 based on pricing, calling workflow, compliance, and AI features.

A sponsored Pharmaceutical Commerce piece by Cryoport Systems argues that biopharma executives should address operational readiness and commercialization planning early, with a downloadable resource and learning objectives.

Sikich announced that its Gov Tech practice leader Karlene Stecchi was named a 2027 Engage Fed Health 150 honoree for her work on federal government technology modernization.

IBM launched a beta ISO 20022 messaging adapter that lets banks instruct tokenized deposit transactions on Swift's permissioned Hyperledger Besu blockchain using their existing cross-border payment message formats, alongside an on-premises deployment of its Digital Asset Haven platform.