# Horizon AI Briefing — 2026-09-16

## Top stories
- **Google Launches Gemini 3.8 Live and Extended Thinking Voice Models, Undercuts OpenAI on Price** (https://the-decoder.com/google-launches-gemini-3-8-live-to-take-on-openais-gpt-live-1-at-a-fraction-of-the-cost/) — Google DeepMind released Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, voice AI models that perform background reasoning and tool execution without breaking conversation flow. Per Google's own claims, the models top the Artificial Analysis speech-to-speech leaderboard; independent reporting confirms pricing at $1.38/hour — a significant undercut of OpenAI's GPT-Live-1 — making competitive real-time voice AI substantially more accessible to developers.
- **OpenAI Pursues Funding at $120B Valuation, a 40% Jump** (https://www.aitimes.com/news/articleView.html?idxno=215324) — OpenAI is seeking new private investment at a $120 billion valuation, up 40% from its prior round, with discussions initiated by existing investors rather than the company itself. The raise signals sustained investor appetite even as OpenAI's IPO timeline remains uncertain, and sets a new benchmark for AI company valuations.
- **Samsung Leads Mistral's €3B Series D at €21B Valuation, Largest-Ever European Tech Round** (https://www.aitimes.com/news/articleView.html?idxno=215321) — Samsung led Mistral AI's €3 billion Series D, valuing the French startup at over €21 billion in what is reported as the largest venture round for a European tech company. The deal pairs a strategic industrial partnership — deploying Mistral Large for semiconductor defect detection and yield optimization — with the capital injection, signaling deepening Asia-Europe AI supply chain ties. [company-reported figures; no independent confirmation]
- **Trump Convenes AI Executives at White House to Navigate Safety vs. Speed Tension** (https://www.aitimes.com/news/articleView.html?idxno=215329) — President Trump is summoning major AI company leaders to the White House, with House Speaker Mike Johnson announcing the meeting will occur within a week, as the administration seeks to balance AI development speed against safety concerns while countering China. The meeting reflects intensifying political pressure on the industry and follows Johnson's public statement that AI companies can self-regulate.
- **Google DeepMind Researcher Resigns, Warns AI Risks Are Accelerating** (https://www.aitimes.com/news/articleView.html?idxno=215325) — Bilal Chughtai, an AI safety and alignment researcher at Google DeepMind, publicly resigned citing serious concerns about the fundamental direction of AI development and the potential for catastrophic outcomes. His departure adds to a pattern of safety-focused researchers leaving major labs and escalating public warnings, increasing scrutiny on lab governance. [self-reported; no independent confirmation]
- **OpenAI Confirms Safety Discussions with Anthropic and Google DeepMind** (https://www.aitimes.com/news/articleView.html?idxno=215328) — OpenAI's global policy chief Chris Lehane confirmed ongoing trilateral safety discussions with Anthropic and Google DeepMind, stating no antitrust exemption is required for such coordination. The disclosure comes as Congress debates whether to grant antitrust cover for safety collaboration and raises questions about what substantive commitments may emerge. [self-reported; no independent confirmation · carried by 2 publishers]
- **Rare Bipartisan Consensus Emerges Around AI Slowdown Calls** (https://www.etnews.com/20260916000011) — At a Future of Life Institute event in Washington, Senator Bernie Sanders and former Trump strategist Steve Bannon publicly aligned on calls to slow AI development, representing an unusual left-right convergence on AI risk. Separately, new NYT/Siena polling shows 61% of likely voters oppose data center construction for AI, underscoring growing public resistance that politicians across the spectrum are responding to.
- **Xiaomi Open-Sources 38B Embodied World Model Robotics-U0** (https://pandaily.com/xiaomi-robotics-u0-open-source-embodied-world-model) — Xiaomi open-sourced Xiaomi-Robotics-U0, an autoregressive embodied world foundation model available in ~4B and ~38B parameter sizes, along with training and inference tooling. Xiaomi claims FlashAR+ inference speedups of ~83x and a top WorldArena ranking for robot-centric scene understanding; the release marks a significant open-source contribution to embodied AI from a consumer electronics giant. [vendor-claimed benchmark; no independent eval]
- **Jensen Huang Pushes Back on AI Regulation, Calls Safety an Engineering Problem** (https://techcrunch.com/2026/09/15/we-dont-need-ai-regulation-leave-safety-to-us-nvidias-jensen-huang-says/) — Nvidia CEO Jensen Huang argued publicly that AI does not require government regulation, characterizing it as hardware and software whose safety can be managed by individual product makers rather than a federal agency. His stance directly counters calls from academics and legislators for an AI regulatory body, and carries weight given Nvidia's central role in the AI supply chain. [self-reported; no independent confirmation]

## Emerging signals
- **Political Momentum Building for a Dedicated U.S. AI Regulatory Agency** (https://bsky.app/profile/pivotpod.bsky.social/post/3mvl7lbtaf22d) — Calls for an 'Artificial Intelligence Agency' modeled on the FAA or FDA are gaining traction across the political spectrum, with academics and commentators framing self-regulation as insufficient for a trillion-dollar industry. The White House meeting and bipartisan slowdown calls suggest the regulatory vacuum may soon be addressed legislatively.
- **Public Opposition to AI Data Centers Reaching Political Inflection Point** (https://www.theverge.com/ai-artificial-intelligence/995917/data-center-nyt-midterm-poll-september) — NYT/Siena polling shows 61% voter opposition to AI data center construction, and local resistance is spreading to cities like Philadelphia with industrial-harm histories. Politicians are starting to respond, suggesting siting and permitting battles could become a material bottleneck for AI infrastructure buildout.
- **On-Device AI Models Gaining Rapid Consumer Traction** (https://www.aitimes.com/news/articleView.html?idxno=215322) — Beydraft's Pocket-35B reached 1.07 million downloads on Hugging Face in seven weeks, averaging nearly 20,000 downloads per day — signaling strong and growing developer and consumer appetite for capable models that run locally without cloud dependency.
- **AI Safety Researcher Exodus from Major Labs Accelerating** (https://bsky.app/profile/bloomberg.com/post/3mvl4c6oftn2c) — The resignation of Bilal Chughtai from Google DeepMind, combined with Jack Clark's public warnings at Anthropic, reflects a pattern of safety-focused insiders leaving or speaking out — a signal that internal alignment between safety researchers and lab leadership is fraying at multiple frontier organizations.
- **China Frames Western AI Slowdown Calls as Geopolitical Strategy** (https://www.aitimes.com/news/articleView.html?idxno=215269) — China's government and state media publicly reframed Anthropic's and OpenAI's safety-based slowdown proposals as Cold War monopoly tactics rather than genuine safety concerns, hardening the geopolitical dimension of AI governance debates ahead of a possible US-China AI dialogue.

## New entrants
- **Gemini 3.8 Live / Gemini 3.8 Live Extended Thinking** (model) — Google DeepMind's new real-time voice AI models featuring background tool execution and multi-step reasoning without interrupting conversation flow; available via consumer products, Workspace, and API with audio watermarking.
- **Xiaomi-Robotics-U0** (model) — Xiaomi's open-source autoregressive embodied world foundation model in 4B and 38B variants, released with full training and inference tooling and claiming top WorldArena rankings for robot-centric scene understanding.
- **Cog-WM 1.0** (model) — EBKernel's JEPA-style latent world model for embodied AI, claiming to raise map-free ObjectNav success to 86.89% and outperform pi0.5 on manipulation benchmarks; launched at the Pujiang Innovation Forum.
- **Meta Luna** (tool) — Meta's camera-free smart glasses codenamed 'Luna,' designed for voice-based AI interaction without a camera to address wearable privacy concerns; slated for production starting October with a likely Meta Connect announcement.
- **Multi-Sigma-Architect** (tool) — Aizos's no-code extension that automatically builds surrogate models from CAE simulation results to streamline engineering design optimization, released as part of the Multi-Sigma platform.

## Biggest movers this week
- **Anthropic** (company) — 609 mentions this week, ↑348 vs the prior week
- **Claude** (model) — 252 mentions this week, ↑115 vs the prior week
- **Dario Amodei** (person) — 76 mentions this week, ↑73 vs the prior week
- **Sam Altman** (person) — 67 mentions this week, ↑49 vs the prior week
- **OpenAI** (company) — 475 mentions this week, ↑42 vs the prior week
- **DeepSeek** (company) — 53 mentions this week, ↑33 vs the prior week

## China & East-Asia AI
- **Xiaomi Open-Sources Robotics-U0 Embodied World Model and Training Stack** (https://pandaily.com/xiaomi-robotics-u0-open-source-embodied-world-model) — rss
- **EBKernel Unveils Cog-WM 1.0 Brain-Inspired Cognitive World Model** (https://pandaily.com/ebkernel-cog-wm-1-0-cognitive-world-model) — rss
- **US government is using a Qwen embedding model for RAG lookup** (https://reddit.com/r/singularity/comments/1wh6g32/us_government_is_using_a_qwen_embedding_model_for/) — reddit
- **Chinese Researchers Unveil Five-Stage Framework for Self-Evolving AI..."Currently at Stage 2"** (https://www.aitimes.com/news/articleView.html?idxno=215284) — rss
- **CrofAI "cheapest inference provider in the world" gets exposed as an OpenRouter wrapper, routing requests to smaller, cheaper models at up to 20x markup. CrofAI responds to Wire Fraud allegations by denying everything, then backtracking, then 3 hours later wiping their entire online presence** (https://reddit.com/r/LocalLLaMA/comments/1wgwe4n/crofai_cheapest_inference_provider_in_the_world/) — reddit

## Korea AI
- **Google Unveils Gemini 3.8 Live, a Voice AI That Thinks While Speaking — 'Top-Rated Voice Quality'** (https://www.aitimes.com/news/articleView.html?idxno=215332) — rss
- **Meta Unveils Camera-Free Smart Glasses to Address Privacy Concerns** (https://www.aitimes.com/news/articleView.html?idxno=215336) — rss
- **Forty2Mar Obtains Government AI Certification for LLM42 and RAG42** (https://www.aitimes.com/news/articleView.html?idxno=215338) — rss
- **Trump Summons AI Executives to White House Amid Clash Over Development Speed vs. Tech Dominance** (https://www.aitimes.com/news/articleView.html?idxno=215329) — rss
- **"Supports New Document Formats Too"... Korea Deep Learning Renews DeepAgent Document AI Solution** (https://www.aitimes.com/news/articleView.html?idxno=215330) — rss

## Japan AI
- **Weathernews Solves "External AI Leakage Concerns" with Proprietary AI — What 80% of Employees Are Using** (https://atmarkit.itmedia.co.jp/ait/articles/2609/16/news030.html) — rss
- **Google Releases Gemini 3.8 Live: Voice Model That Executes Tasks in the Background Without Interrupting Conversation, Plus Extended Thinking Variant** (https://www.itmedia.co.jp/news/article/2609/16/2000001524/) — rss
- **Where Should We Run AI In-House? Hearing from HP Japan's Practitioner on Cloud vs. Local Deployment** (https://www.itmedia.co.jp/aiplus/article/2609/16/2000001383/) — rss
- **Aizos Releases Multi-Sigma-Architect: Automatic Surrogate Model Construction from CAE Results for No-Code Optimization** (https://monoist.itmedia.co.jp/mn/articles/2609/16/news023.html) — rss
- **"The shortage isn't COBOL talent" — Hitachi on the role of people in AI-era system modernization** (https://www.itmedia.co.jp/enterprise/articles/2609/16/news012.html) — rss

## Europe (EU) AI
- **Why Samsung Paid for Mistral When the Model Is Free to Use** (https://www.aitimes.com/news/articleView.html?idxno=215321) — rss
- **Google launches Gemini 3.8 Live to take on OpenAI's GPT-Live-1 at a fraction of the cost** (https://the-decoder.com/google-launches-gemini-3-8-live-to-take-on-openais-gpt-live-1-at-a-fraction-of-the-cost/) — rss
- **AI labs have a data trust problem that their policies haven't solved** (https://the-decoder.com/ai-labs-have-a-data-trust-problem-that-their-policies-havent-solved/) — rss
- **After warning AI is too dangerous, Bill Gates bets a billion on its upside** (https://the-decoder.com/after-warning-ai-is-too-dangerous-bill-gates-bets-a-billion-on-its-upside/) — rss
- **Exein raises $270m to fight AI hackers, doubles valuation** (https://sifted.eu/articles/exein-funding-round-ai-hack-cybersecurity/) — rss

## Regulation updates
- [🇺🇸 US] **To better coordinate Federal efforts to utilize advance technologies, such as artificial intelligence, to improve diagnoses, treatments, cures, and prevention strategies for pediatric cancer, and for other purposes.** — Proposed. Introduced in House
- [🇺🇸 US] **To provide for certain artificial intelligence agent discovery and security standards, and for other purposes.** — Proposed. Referred to the Committee on Science, Space, and Technology, and in addition to the Committee on Oversight and Government Reform, for a period to be subsequently determined by the Speaker, in each case for consideration of such provisions as fall within the jurisdiction of the committee concerned.
- [🇺🇸 State] **Prohibits private use and regulates government use of automated license plate readers.** — Proposed. Tracked
- [🇺🇸 State] **Limits retention period for automated license plate reader data.** — Proposed. Tracked
- [🇺🇸 State] **Prohibits certain use of artificial intelligence in denying homeowner, automobile, and flood insurance claims.** — Proposed. Tracked
- [🇺🇸 State] **Regulates use of artificial intelligence in certain mortgage lending tools to minimize discrimination in lending.** — Proposed. Tracked
- [🇺🇸 State] **Establishes six month moratorium on approvals, construction, expansion, subsidies, and tax incentives for certain artificial intelligence data centers; creates Artificial Intelligence Data Center Impact Commission.** — Proposed. Tracked
- [🇺🇸 State] **Enacts the "utility billing integrity act" to establish utility billing integrity and consumer protections through anomaly detection, advanced data analytics and the usage of artificial intelligence; requires every utility to implement and maintain a billing integrity program utilizing anomaly detection systems to review all residential utility bills prior to issuance; provides that where a billing anomaly is identified, the utility shall provide prompt notice to the customer that the bill is under review.** — Proposed. Tracked
- [🇺🇸 US] **Research and Oversight of AI in Courts Act of 2026** — Proposed. Read twice and referred to the Committee on the Judiciary.
- [🇺🇸 State] **AI SAFETY MEASURES ACT** — Proposed. Tracked
- [🇺🇸 State] **AN ACT to amend Tennessee Code Annotated, Title 4; Title 10; Title 16; Title 36; Title 37; Title 38; Title 39; Title 40; Title 45; Title 47; Title 48; Title 49; Title 53; Title 63; Title 67; Title 68 and Title 71, relative to artificial intelligence.** — Proposed. Tracked
- [🇺🇸 US] **AI for Mainstreet Act** — Proposed. Introduced in Senate
- [🇺🇸 State] **California Preventing Algorithmic Collusion Act of 2025.** — Floor Action. Tracked

---
Source: Horizon (https://horizon.alchemylab.sh) — aggregated, LLM-scored AI intelligence; each item also lists its own primary source. Cite both — a ready-to-paste citation is in provider.citation.