AI Week, Nov 2–8, 2026: Gulf Deals, a 2.8T Open Model and an Nvidia Board Inquiry
OpenAI signed four agreements in Abu Dhabi, Moonshot opened the Kimi K3.1 Pro weights, Google made Gemini Omni 1.2 Flash generally available, and Nvidia's board opened an independent investigation involving CEO Jensen Huang.
Written on November 13, 2026, covering news from Nov 2–8, 2026.
The first full week of November had 2 threads: AI labs moving deeper into government and state-company deployments, and a run of releases that put very large open weights and cheaper AI video on the table. It closed with Nvidia’s board opening an independent investigation tied to its CEO.
Monday, November 2
In a November 2 roundup of what Apple still has planned for 2026, MacRumors reported that Apple has confirmed a December public rollout of its first Apple Intelligence features for users in China, shipping alongside iOS 27.3. Those features are already in testing in the iOS 27.3 beta, and the same roundup says the first beta of that release arrived on Tuesday, November 3. Apple has partnered with Alibaba to bring Qwen support to Siri, Writing Tools and Visual Intelligence. The first release covers supported iPhones only, other platforms will follow after the required regulatory approvals, and Siri AI is not included. Apple has said iOS 27.3 reaches the public in December but has not given a date.
In Abu Dhabi, OpenAI chief executive Sam Altman met Sheikh Khaled bin Mohamed, Crown Prince of Abu Dhabi and Chairman of the Abu Dhabi Executive Council, and a set of agreements followed. AI Weekly, citing The National, counted 4 deals signed on Monday at ADIPEC, with the Crown Prince and Altman witnessing the ceremony.
- ADNOC: The state energy company signed a multi-year partnership with OpenAI and G42, through G42’s subsidiary Inception42, to deploy OpenAI’s Frontier platform; AI Weekly says this makes it the first energy company to do so. Initial work spans field planning, production, processing, refining, logistics, sales and trading, with systems on UAE-based infrastructure and human experts keeping oversight of consequential decisions.
- Machine-Readable Government: Led and governed by the Department of Government Enablement – Abu Dhabi (DGE), the initiative converts a defined set of policies and regulations into machine-readable form so AI systems can help officials with routine compliance checks, such as those for permits and certificates. Officials still own final decisions, and a live pilot covering food imports is running with the Abu Dhabi Agriculture and Food Safety Authority (Adafsa). It runs on OpenAI’s GPT models with inference infrastructure based in the UAE.
- Investment analysis: Mubadala, Inception42 and OpenAI agreed to apply AI to investment analysis, opportunity identification and risk assessment.
- 5-year Frontier rollout: Inception42 and OpenAI signed a five-year agreement covering adoption of Frontier across government, enterprise and education.
AI Weekly reported that no contract values were disclosed. Altman said in a statement that Abu Dhabi and the UAE are “setting a bold, global standard for AI integration across both the public and private sectors”.
CNBC placed the deals alongside OpenAI’s effort to raise $20 billion in a new round co-led by UAE investors, which it reported on Saturday. CNBC also noted that MGX, an Abu Dhabi investment firm, co-led OpenAI’s $122 billion raise in March and Anthropic’s $30 billion raise in February.
Wednesday, November 4
Moonshot AI announced on X on November 4 that it was releasing the weights of Kimi K3.1 Pro, now hosted on Hugging Face as moonshotai/Kimi-K3.1-Pro. AILog notes the tracker AI/TLDR logged the upload on November 2, inside the November 4 deadline Moonshot set when the model launched on October 23.
The model card describes a 2.8-trillion-parameter Mixture-of-Experts model with 104 billion parameters active per token, a 1,048,576-token context window and image input via the MoonViT-V2 encoder. The router picks 16 of 896 experts for every token, alongside 2 shared ones, over 93 layers; weights ship in MXFP4 with MXFP8 activations thanks to quantization-aware training. i10x puts the download at about 1.56 TB across 96 shards. An OpenAI/Anthropic-compatible API is also available at platform.kimi.ai under the id kimi-k3.1-pro.
The Kimi K3.1 Pro License allows commercial use, but a Model-as-a-Service business whose aggregate revenue, counting affiliates, tops $20M over any consecutive 12 months must sign a separate agreement with Moonshot first. Products above 100 million monthly active users, or above $20M in monthly revenue, must prominently display “Kimi K3.1-Pro”. Both clauses carry exemptions: internal use, plus use via Moonshot’s own official products or its certified inference partners. On Moonshot’s own table the model scores 65.7 on Terminal-Bench 4.0, just under Claude Opus 5.5 at 66.4 and ahead of GPT-6 Astra at 58.2; all of these scores are self-reported.
Thursday, November 5
Google began rolling out Gemini Omni 1.2 Flash and made it generally available across the Gemini API and Gemini for business. Clips can now run natively up to 15 seconds, up from 10 seconds on Omni 1.1, with length set in one-second steps from 3 to 15 seconds. Google cut the price 30% from Gemini Omni 1.1, to $0.07 per second of 720p video.
Other changes include multi-turn conversational editing, closer adherence to image, short motion-clip and audio references, and drafting at 360p or 720p before upscaling to 1080p or 4K. Outputs carry an imperceptible SynthID watermark, and Provisioned Throughput is offered. In the Gemini app, 15-second video is rolling out to Google AI Ultra subscribers. Google said development continues on harder cases such as fast sports action, rapid script changes and complex physics.
On November 6, Google Cloud announced general availability on its platform, noting that audio references are gated behind allowlisting and listing Gemini for business as coming soon. Photoroom CEO Matt Rouif said in that post: “Version 1.2 brings 15-second videos, a top user request.”
Friday, November 6
Manaknight Digital’s weekly roundup, prepared November 6, highlighted an assessment that the AI evaluation nonprofit METR published on November 5 about Anthropic. METR based it on an 11-week review of Anthropic’s internal research data plus staff interviews. It estimated that Anthropic’s model capabilities on well-defined benchmark tasks are improving at around 1.9 times the pace seen before October 2025.
The researchers tied much of that speed-up to Anthropic using AI tools in its own research and development, rather than to faster growth in training compute. METR flagged the estimate as highly uncertain and said AI is not yet doing major work on its own in the most important parts of training. METR also said Anthropic redacted a large share of the evidence it found most useful.
Saturday, November 7
Daily AI roundups dated November 7 summarised a series of upgrades to Meta’s Muse assistant. According to AIdapted, Meta launched Muse Realtime Voice and Muse Realtime Avatar on November 5, giving users in the US real-time voice conversations with Muse. The avatar reacts with expressions and gestures synced to speech, and users can cut in while the model is answering.
AI Agents Directory adds that the voice-and-avatar feature covers mobile, web and Mac, and that Meta introduced ‘Spaces’ to help organise ongoing projects. AIdapted reports that on November 6 Meta also announced specialised models for Muse and Meta AI built on Muse Wonder technology, with Muse’s score on the Personal Superintelligence Index rising from 38% to 47%.
Sunday, November 8
Nvidia’s board said it is opening an independent investigation following reports that a foreign intelligence effort targeted founder and CEO Jensen Huang. Independent directors will run it as a committee, with help from outside independent counsel. CNBC reported that Huang wrote to staff by email on Sunday, telling them the board had opened an independent investigation into the allegations.
The board stated: “While the Board has no reason to believe that either national security or NVIDIA secrets were compromised, the Board’s inquiry will examine all available information.” It said Huang fully supports the review and has committed to cooperating, and that it has confidence in him and his leadership team. An Nvidia spokesperson said Huang was committed to continuing as CEO without interruption, according to CNBC.
CNBC noted the backdrop: U.S. rules curb China sales of Nvidia’s most advanced AI chips, while Huang has argued for wider AI chip exports there.
Also this week
Reported on November 5: Digital Trends covered OpenAI’s Images 2.6 update, which adds GIF support to ChatGPT, Codex and the API. According to OpenAI, image generation is 50% quicker than the previous model, and characters, products and brand elements stay consistent across edits. Developers get 2 new models: GPT-Image-2.6 Flare, tuned for speed, which can finish an image in six seconds at best on low quality, and GPT-Image-2.6 Sunburst, focused on detail. The API also adds image markup, SVG export and a complexity setting. By OpenAI’s count, users create roughly 4 billion images weekly across these platforms. Manaknight Digital dates the Images 2.6 introduction to November 4.
Reported on November 3: Anthropic’s amended S-1 IPO filing warns investors that the company “may imminently meet” its own Responsible Scaling Policy definition of “dramatically accelerated progress attributable to AI,” its tripwire for AI that speeds up the building of the next AI. According to the filing, that view comes from internal estimates and expectations about future models, not its current ones. Anthropic says it has not designated any current model as exceeding the threshold. The amendment also turns an earlier forecast into an admission: the company has “in the past not identified, and may in the future not identify” a significant speedup in its development pace until after the fact. Anthropic says it does not anticipate being able to safely have its AI design and build successive frontier generations fully autonomously without meaningful human supervision, and it is considering policy updates “to keep humans in control of frontier model development at Anthropic”. It plans to list on Nasdaq under the ticker ANT.
What to Watch
- Nvidia’s next 2 weeks. Nvidia reports third-quarter earnings on Nov. 17, and 3 House Republican committee chairs have written to Huang requesting a closed-door briefing for lawmakers on Nov. 18. The board has said it will give further updates as appropriate; I’d watch whether the first one comes before or after earnings.
- Apple in China in December. The checkable point is whether the Qwen-backed features land with iOS 27.3 in December and which further platforms clear regulatory approval after that.
- OpenAI’s $20 billion round. CNBC reported the round is co-led by UAE investors; my read is that the Abu Dhabi signings make the round’s final investor list the next thing worth checking.
- Kimi K3.1 Pro under independent testing. i10x asks whether independent labs will reproduce Moonshot’s coding and agent scores and whether a fuller K3.1 technical report follows the model card.
This article was written by AI from the sources listed on this page. Figures, names and quotes were checked against those sources by software before publishing.