AI News Feed

01 Oct 2025- Four major AI launches: OpenAI adds in-chat Instant Checkout with Stripe; Anthropic releases Claude Sonnet 4.5 and developer tools; Microsoft adds Copilot Agent Mode; DeepSeek unveils cost‑saving long‑context model.

30 Sep 2025- Microsoft centralizes Windows engineering under one organization led by Pavan Davuluri, consolidating core teams to accelerate AI features and create an "Agentic OS," while some components remain with Azure.

30 Sep 2025- California enacted SB 53, requiring large "frontier AI" developers to publish safety/security frameworks and updates, establish incident reporting and whistleblower protections, with civil penalties; industry reaction split.

30 Sep 2025- Xicoia’s synthetic performer Tilly Norwood is courting Hollywood agencies, sparking actor boycotts and industry debate over representation, compensation, and ownership of AI-driven performers.

29 Sep 2025- Generative-AI firms court Hollywood with text-to-image/video deals (e.g., Critterz), promising cost cuts but facing technical limits, legal battles, labor concerns, and uncertain creative value.

29 Sep 2025- Trump posted then removed an AI-generated fake Fox News clip on Truth Social promoting the “MedBed” conspiracy; Fox News says the segment never aired and the motive is unclear.

29 Sep 2025- Larry Ellison and Oracle expanding into media — reportedly seeking US control of TikTok — raising concerns about consolidation, influence over news, and combining data, AI infrastructure, and platforms.

29 Sep 2025- Turing winner Richard Sutton argues LLMs mimic language without goals; advocates continual, reward-driven OaK agents that learn from experience, highlighting autonomy versus safety amid rapid compute scaling.

27 Sep 2025- YouTube is testing AI hosts in YouTube Music radio stations and mixes via a limited YouTube Labs trial, offering commentary, stories, and trivia to enhance listening for select U.S. participants.

27 Sep 2025- Meta launches internal robotics effort to build humanoid "Metabot," pairing with its Superintelligence AI lab to develop world models and dexterous manipulation software, aiming to license robot software to manufacturers.

27 Sep 2025- Microsoft tests AI in Windows 11 Photos to auto-sort images into receipts, screenshots, identity documents, and handwritten notes (cross-language), rolling out to Insiders on Copilot Plus PCs.

27 Sep 2025- OpenAI launched ChatGPT Pulse, a Pro-preview that generates personalized daily morning briefings from chat history and optional Gmail/Calendar, producing 5–10 curated cards with feedback controls and usage caps.

27 Sep 2025- OpenAI's GDPval benchmark tests 1,320 real‑world tasks across 44 occupations; Claude Opus 4.1 often outperformed humans, GPT‑5 excelled in accuracy, showing huge speed/cost gains but persistent errors and risks.

26 Sep 2025- Microsoft prototyped AI‑designed microfluidic cooling etched into silicon channels, cutting GPU silicon temperature rise 65% and promising denser, more efficient datacenters, though timeline and durability remain uncertain.

26 Sep 2025- Google Cloud’s DORA report: ~90% of developers use AI assistants (~2 hours/day), improving efficiency and code quality but trust lags; DORA AI Capabilities Model outlines seven practices.

26 Sep 2025- Microsoft adds Anthropic’s Claude (Sonnet 4, Opus 4.1) to Microsoft 365 Copilot, creating an opt‑in multi‑model picker; Claude runs on Anthropic’s infrastructure, sending data outside Microsoft.

25 Sep 2025- Qualcomm unveiled the Snapdragon 8 Elite Gen 5: a flagship mobile SoC emphasizing on‑device AI with a 3rd‑gen Oryon CPU, faster Hexagon NPU, improved GPU, camera AI and X85 modem.

25 Sep 2025- Nvidia and OpenAI signed a LOI for Nvidia to deploy 10 GW of systems (millions of GPUs) and potentially invest up to $100B; first gigawatt via Vera Rubin in 2026.

25 Sep 2025- OpenAI, Oracle and SoftBank unveil a massive "Stargate" data‑center buildout targeting ~7 GW and $400B in investments; Altman touts "Abundant Intelligence"; voice-privacy demo and Alibaba Qwen updates.

22 Sep 2025- Trump's new H‑1B proclamation hikes visa costs to $100,000 and creates a broad DHS "national interest" waiver, enabling political leverage over tech firms, banks, and universities.

22 Sep 2025- OpenAI is recruiting ex‑Apple hardware talent and partnering with iPhone suppliers to build consumer AI devices—speaker, glasses, wearables—targeting a late‑2026/early‑2027 debut.

22 Sep 2025- Microsoft and xAI race to build gigawatt‑scale AI datacenters—massive GPU fleets, power and water solutions—while Grok model releases and infrastructure control shape AI’s near future.

19 Sep 2025- Meta CTO Bosworth attributes Ray‑Ban smart glasses demo failures to dev‑server routing that triggered Live AI across devices (a self‑inflicted DDoS) and a separate sleep‑state call bug, now fixed.

19 Sep 2025- Radware demonstrated "Shadow Leak" — a prompt-injection attack that made a ChatGPT agent exfiltrate Gmail data via a hidden email prompt; OpenAI patched it, warning risks to other connectors.

19 Sep 2025- Google integrated Gemini into Chrome for all U.S. desktop users, adding a sidebar for cross-tab AI assistance, an upcoming address-bar "AI Mode," and previews of autonomous multi-step agent features.

18 Sep 2025- Reddit is renegotiating licensing with Google and reportedly OpenAI, seeking higher, usage‑based payments and support to restore traffic after AI tools surface Reddit content and reduce engagement.

18 Sep 2025- Nvidia invests $5B in Intel to co-develop x86 SoCs combining Intel CPUs with Nvidia RTX GPU chiplets, using NVLink to build custom AI/data-center and PC products; Intel stock jumps.

18 Sep 2025- Meta unveiled Ray‑Ban and Oakley smart glasses plus a Neural Band that decodes subtle muscle signals for discreet, intent-driven controls; Ray‑Ban Meta Gen2 offers ~8‑hour battery and 3K video.

18 Sep 2025- OpenAI’s ensemble reasoning models, including GPT‑5, solved all 12 ICPC World Finals problems under contest rules, beating top human teams; Google’s Gemini 2.5 solved 10/12.

17 Sep 2025- Google tests a Windows Spotlight‑like Search app via Search Labs: Alt+Space summons a resizable bar searching PC files, Drive, web, Lens, and optional AI Mode; US English, Windows 10+.

17 Sep 2025- OpenAI will alter ChatGPT for under-18s: age‑prediction/ID checks, teen‑safe defaults (no flirtation or suicide discussion), parental controls/notifications, contacting parents/authorities if imminent risk.

17 Sep 2025- DeepMind CEO Demis Hassabis urges lifelong "learning how to learn" as AI — possibly AGI within a decade — reshapes jobs and education, requiring continual upskilling.

17 Sep 2025- OpenAI and Anthropic usage reports show Claude users focus on coding, ChatGPT on writing and advice, rising personal use (53%→73%) and rapid growth in low/middle‑income countries, and increasing information‑seeking/task delegation.

17 Sep 2025- MIT spin‑out AlterEgo unveiled a "telepathic" wearable that reads neural speech‑intent signals to map to words, targeting communication restoration for ALS/MS patients; early demos report ~92% accuracy.

17 Sep 2025- OpenAI launches GPT-5 Codex, a coding agent that autonomously works for hours with faster throughput and automated reviews. Usage grew 10x; companies like Duolingo and Cisco Meraki deploy it.

15 Sep 2025- Penske Media, publisher of Rolling Stone and The Hollywood Reporter, sued Google, alleging its AI Overviews unlawfully summarize reporting, siphoning traffic and revenue from publishers.

15 Sep 2025- Apple adds real‑time on‑device AirPods translation powered by Apple Intelligence—offline audio processing, five languages at launch; replies show as text on other iPhone. One‑sided unless both devices compatible; EU blocked.

13 Sep 2025- Encyclopedia Britannica and Merriam‑Webster sued AI search Perplexity, alleging copyright and trademark infringement for scraping, republishing definitions, and diverting web traffic without permission.

13 Sep 2025- Tucker Carlson accused OpenAI's Sam Altman of involvement in Suchir Balaji’s death, despite San Francisco ruling it a suicide; Altman denied wrongdoing and cited police reports.

12 Sep 2025- Users applied AI "upscaling" to the FBI’s blurry Charlie Kirk suspect photos, generating speculative, often inaccurate faces that shouldn’t be treated as evidence or definitive identification.

OpenAI backs AI-driven animated feature Critterz — made mainly with GPT-5 — aiming 2026 Cannes/theatrical release; produced under $30M in nine months to showcase cheaper, faster AI filmmaking amid industry resistance.

Anthropic’s Claude can now create and edit real files (Excel, Docs, PowerPoint, PDFs) in Claude.ai/desktop—running code in a private environment. Preview available; internet access raises data-risk warnings.

Firefox adds "shake to summarize" on iPhone, using Apple Intelligence on-device (iPhone 15 Pro+/iOS 26) or Mozilla cloud fallback; English US launch, optional, 5,000-word limit.

Google updates Veo 3 and Veo 3 Fast: API adds 9:16 vertical video and up to 1080p, lowers prices, and targets developers to boost AI-generated social clips.

Judge Alsup paused Anthropic’s $1.5B settlement with authors over AI training-piracy claims, citing secretive negotiations and an unclear claims process; approval to be reconsidered at Sept. 25 hearing.

Grammarly expands its AI writing tools to Spanish, French, Portuguese, German and Italian, adds translations to 19 languages, using fine-tuned open-source LLMs with optional third-party models.

Ant Group's Robbyant showcased R1 humanoid at IFA and Shanghai demos, performing slow kitchen tasks; pitched for caregiving, tourism and service roles, no launch date or pricing announced.

Microsoft bundles Sales, Service, and Finance Copilots into Microsoft 365 Copilot from October, cutting bundled cost to $30/user, adding Agent 365 management tools, and partly using Anthropic models.

Users posted AI-upscaled versions of the FBI’s blurry photos of the Charlie Kirk shooting suspect; generative tools fabricate plausible but unreliable details, so such upscales aren’t trustworthy evidence.

Anthropic adds automatic, optional memory to Claude (Team/Enterprise), linking memories to projects for context-aware outputs; also launches incognito chats amid concerns about long-term memory risks.