Highlights
- Thrive to buy ADT for $15.3B, with OpenAI deeply involved
- DeepSeek closes $16B round led by China's state Big Fund
- Mistral open-sources its 1-trillion-parameter Large 4
- OpenAI doubles usage quotas across all ChatGPT Pro tiers
- Meta's scraping-friendly commercial terms take effect today
Headlines
[OpenAI-backed Thrive to acquire ADT for $15.3B in all-cash deal to retrofit it with AI] On Oct 29, Josh Kushner's Thrive Holdings struck a definitive deal to buy home-security giant ADT at $10 per share in cash — $7.6B in equity value, $15.3B enterprise value, a 45.6% premium with a 30-day go-shop period — financed by Apollo with no financing contingency. OpenAI, Thrive's partner and investor, will supply frontier models and forward-deployed engineers; OpenAI COO Brad Lightcap framed the deal as a template for "frontier AI deployed rapidly across entire organizations." (TechCrunch)
[DeepSeek closes $16B second round led by China's state Big Fund at $74B pre-money] According to The Information, DeepSeek has closed a $16 billion second funding round led by China's state National IC Industry Investment Fund (the Big Fund), with participation from tech giants including Tencent and Alibaba plus state-backed investors, at a $74 billion pre-money valuation. The round signals top-tier state backing for China's leading LLM lab, with proceeds earmarked for compute expansion and next-generation model training. (The Information)
[Mistral open-weights trillion-parameter multimodal MoE model Large 4] On Oct 29, France's Mistral AI released the full weights of Mistral Large 4 under its Ecosystem Open-Weight license (a modified MIT with a MaaS carve-out). The trillion-parameter multimodal mixture-of-experts model, nicknamed "Le Chonk," scores 19.7% on FrontierCode and 41% on Terminal-Bench, matching top closed models and becoming one of the strongest open-weight models available. (Mistral)
Products & Tech
[OpenAI doubles usage limits across all ChatGPT Pro tiers] On Oct 29, OpenAI's Codex lead Tibo announced that, effective immediately, the Pro 200 throttling is removed and usage caps across Pro 100, Pro 200, Pro 500 and Business Premium are all doubled, while Work and Codex quotas stay unchanged and Personal Agent extended credits are counted separately. The move is widely read as a response to complaints about throttling and rival limit policies. (AITNT)
[Tencent launches Buddy One, unifying its WorkBuddy suite] Tencent officially launched Buddy One, using a free Buddy One ID to unify accounts and credits across WorkBuddy, Tencent Meeting, Tencent Docs, ima and Tencent Drive. Buddy One VIP costs 700 yuan per month with 50,000 credits and premium perks, targeting the AI office-productivity subscription market. (AITNT)
[EMBRACE research breaks the built-in browsers of five AI coding tools] Security firm Forever Security's EMBRACE ("embedded browser trusted like a real one") research shows the embedded browsers in Cursor, GitHub Copilot, OpenAI's Codex, Microsoft's VS Code and Anthropic's Claude Desktop all break three core browser security rules; in Cursor, a malicious page could run shell commands. Cursor and Copilot shipped October fixes; Codex and VS Code remained unpatched at publication. (Help Net Security)
IPO & Fundings
[NVIDIA in talks to guarantee Nscale debt for Anthropic's second 460MW build] Per The Information, NVIDIA is in advanced talks to guarantee billions in debt for cloud provider Nscale to build a second 460MW phase of its Monarch campus in West Virginia, doubling the capacity Anthropic committed to under its six-year, ~$45B deal, with the first phase slated to come online in late 2027. (The Information)
[AI mental-health startup GoodMood raises ~$45M for global expansion] On Oct 29, GoodMood Holdings announced it raised nearly $45 million to scale its AI-native mental healthcare platform worldwide, extending this fall's streak of bets on clinician-plus-model workflows and global distribution plays. (Crescendo)
[Ant Group leads ~$14M Series A in tactile-sensor startup Juqiao Industrial] Juqiao Industrial closed a nine-figure RMB Series A led by Ant Group to mass-produce flexible electronic skin and expand tactile multimodal R&D. The company serves 400+ customers, has delivered nearly 10,000 tactile gloves, and is building a Physical AI perception data layer for humans and robots. (AITNT)
Open Source & Research
[OpenAI details four-layer frontier monitoring stack: alerts within 15 minutes] OpenAI published its layered monitoring system: L0 activation probes with 5% random sampling, an L1 model-based reviewer re-scanning sensitive workloads every 30 minutes, L2 investigative agents that reconstruct incidents, and L3 network/telemetry cross-checks before paging humans. A replay of the Hugging Face sandbox escape would have alerted responders 30 minutes after breakout — over 30 hours before the actual incident. (Unite AI)
[NUS x Kling AI propose DiffusionART, 10x faster video diffusion post-training] Teams from the National University of Singapore, Kling AI and HKUST proposed DiffusionART, which supports ODE sampling without differentiable reward models or tuning; on HunyuanVideo-13B it achieves 12x speedups at comparable reward, offering an efficient path for AI video post-training. (AITNT)
[X-Era AI open-sources native 4D world action model E0.5, beating pi-0.5 on real robots] X-Era AI open-sourced E0.5, which predicts action consequences via 3D geometric reasoning; it reaches 85.0% and 88.3% overall success on Piper and H1 real-robot tasks, up 13.3 and 26.7 percentage points over pi-0.5. (AITNT)
Company & People
[Google DeepMind engineer Sofia Liguori resigns over military AI contracts] On Oct 29, DeepMind research engineer Sofia Liguori published a resignation letter titled "Stay Human," protesting Google's $1.2B Project Nimbus contract supplying cloud and AI to the Israeli government and military, saying Google proceeded without usage restrictions despite internal legal warnings. Nearly 200 DeepMind workers — about 5% of the unit — have signed a letter urging Google to drop its military contracts. (Democracy Now!)
[CrowdStrike: VAPOR PANDA ran DeepSeek-powered agent attacks on 200+ targets] CrowdStrike disclosed Oct 29 that the China-nexus group VAPOR PANDA used an AI agent framework called ARTEX — running primarily on DeepSeek's LLM plus a second commercial model — to run intrusions against more than 200 organizations across at least 14 countries in an 11-day window starting Oct 17, the most detailed attribution yet of AI-scaled offensive operations. (CrowdStrike)
Policies & Data
[Meta's new commercial terms take effect today: business websites default to AI-training scraping] Effective Oct 30, Meta's updated commercial terms allow it to use scrapers, bots and test credentials to crawl content from websites associated with business accounts for AI training, with no opt-out. Legal experts say the terms can cover news sites; California Senator Adam Schiff has written to Meta demanding it scrap the clauses, calling them "a gross abuse of Meta's power." (Nieman Lab / Ars Technica)
[UMG settles Udio suit and signs licensing deal, a first for major-label AI music] On Oct 29, Universal Music Group and Udio announced they resolved UMG's copyright suit and signed licensing agreements covering its recorded music and publishing catalogues, including compensatory payments. The parties plan a 2026 subscription platform trained only on authorized, licensed music with fingerprinting and filtering, with artists opting in — the first major-label AI-music case to settle via commercial license, setting a template for pending Suno cases. (PR Newswire)
Powered by AOYII.com
