This week, politics and money shaped AI more than any single model launch.On the research side, an independent METR study found that Anthropic’s models now improve about twice as fast as before, mostly because AI helps build AI. OpenAI launched GPT-6 Cyber for security researchers and added GIF creation to ChatGPT. Meta, CrowdStrike and the US Justice Department also made moves.
Voters Reject AI Data Centers
On November 3, US voters decided 41 local ballot measures about data centers. In 40 of them, the majority voted for more limits on data center construction. Meta said the backlash shows how people feel about AI in general, and it could make AI expansion harder.
AI Stocks Lose $2.5 Trillion
AI stocks fell sharply after the Democratic sweep in the midterm elections. On November 5, the Dow dropped about 1,800 points and chip stocks had their worst day since March 2020. Nvidia fell about 9%, its worst day since January 2025, then recovered 2.8% on Friday.
- Investors fear new rules could slow the data center buildout.
- South Korea’s Kospi fell as much as 8.8%, which triggered a trading halt.
OpenAI Launches a Community Compact
Two days after the vote, OpenAI published a “National Community Compact” for its data centers. OpenAI will pay the full electricity and infrastructure costs at sites it owns and operates. It will also stop using nondisclosure agreements with public officials and publish local job reports.
METR: AI Doubles Anthropic’s Speed
On November 5, the independent nonprofit METR published an 11-week review of Anthropic’s internal research data. METR found that Anthropic’s models now improve about 1.9 times faster than in 2024–2025. It estimates that about 80% of this speedup comes from Anthropic’s own use of AI in research.
- The estimate has high uncertainty (95% range: 1.5x to 2.8x).
- Anthropic redacted most of the evidence that METR found most informative.
Tencent Releases Hy4
Tencent released Hy4 on November 4, the full version of the model it previewed in August. Hy4 is a 770-billion-parameter mixture-of-experts model, and it is the first Hy model that can understand images and operate a computer. It is available now through Tencent Cloud and OpenRouter, and Tencent plans to release the weights in the coming days.
Anthropic Warns of Self-Improving AI
Anthropic’s revised IPO filing on November 2 says it may soon reach its own threshold for AI that dramatically speeds up its own development. This idea is called recursive self-improvement (RSI). Anthropic says AI that helps build more capable AI can be a safety risk if progress becomes too fast.
OpenAI Launches GPT-6 Cyber
OpenAI released GPT-6 Cyber, a version of GPT-6 Astra trained for authorized security research. Researchers used it to find 81 unknown high-severity bugs in Firefox and build working exploits for 948 Linux kernel bugs. Without safety prompts, it stayed inside the authorized scope in 99.93% of tests.
- OpenAI rates it “Critical,” the top cyber risk level in its Preparedness Framework.
- Access is limited to approved Daybreak Red customers.
ChatGPT Can Now Make GIFs
OpenAI launched Images 2.6 on November 4. Users can now make animated GIFs in ChatGPT from a text prompt or a photo, then change the speed and edit single frames. Developers get two new API models: GPT-Image-2.6 Flare for speed and Sunburst for quality.
OpenAI says users now make more than 4 billion images per week.
Meta Acquires AI Coding Startup Conductor
Meta has acquired Conductor, an AI coding startup founded in 2024 whose tool lets engineers run several AI agents on different parts of a project at once. Meta confirmed the deal but did not share terms. Conductor’s main features will keep working until early next year, but its cloud service shuts down on November 20. The deal is part of Meta’s push to catch up in AI coding, after it launched its own coding agent, Muse Code, in August.
Apple and Goldman Limit AI Tool Use
Apple cut the daily AI spending limit for its engineers from $300 to $200, and some teams may drop to $100, Bloomberg reported. Goldman Sachs now lets only its roughly 12,000 engineers use Claude Code and GitHub Copilot, and it removed access for traders and bankers. Both moves show companies now control AI costs more closely.
ADNOC Partners With OpenAI and G42
On November 2, Abu Dhabi’s state oil company ADNOC signed a multi-year deal with OpenAI and G42. ADNOC becomes the first energy company to use OpenAI Frontier, a platform for building and managing AI agents. The system will run on G42’s AI infrastructure in the UAE.
Google Releases Gemini Omni 1.2 Flash
On November 5, Google made Gemini Omni 1.2 Flash, its newest video generation model, generally available. It can generate and edit clips up to 15 seconds long, up from 10 seconds. 720p video now costs $0.07 per second, which is 30% less than Omni 1.1 Flash.
- Available in the Gemini API, Gemini for business, Flow and Google Vids.
- 15-second videos in the Gemini app are rolling out to Google AI Ultra subscribers.
IndiaAI Mission Opens Procurement for 10,000 Compute Units
C-DAC officially invited bids under the IndiaAI Mission to procure up to 10,000 domestic accelerator units (GPUs and TPUs) to support startups, academic research, and government applications with mandatory domestic data residency.
Thinking Machines Releases Inkling 2 Small
Mira Murati’s Thinking Machines Lab released Inkling 2 Small on November 5, the first model of its second generation. It is a mixture-of-experts model with 594 billion total parameters, and 17 billion are active per token. It is available now on the company’s Tinker platform, and the full weights will follow after more safety testing.