From election-day votes against data centers to new AI tools for video, money and cybersecurity, here are the AI stories that mattered most this week.
Top stories
- AI and data centers became a midterm election issue. Ohio voters rejected data centers across 21 ballot measures, and CNBC reported that voters backed restrictions in 40 of 41 U.S. ballot measures. Democrats, preparing to take control of Congress, promised AI hearings and “reasonable guardrails,” and an NBC News exit poll found more than half of voters called AI and data centers a crucial or major factor in their vote.
- OpenAI promised to pay its own way on data centers. In a new “National Community Compact,” OpenAI said it will pay the full electricity and required infrastructure costs at data centers it owns and operates, and will stop signing nondisclosure agreements with public officials.
- A report says Anthropic is building AI about twice as fast as before. Independent researchers at METR reviewed Anthropic’s internal data from August 12 to October 29 and estimated its pace of progress has roughly doubled. Most of the speedup came from AI helping employees with coding and analysis, not from AI running research on its own; much of the roughly 100-page report was redacted.
- AI stocks slipped as investors questioned the building boom. Shares fell amid doubts that AI infrastructure spending will keep its expected pace, as places from New York to Amsterdam tightened data center rules. Google still announced a $17 billion data center in Wyoming, partly powered by a planned 702-megawatt fuel-cell project.
- The Justice Department launched a “Super Intelligence Task Force.” The new unit will investigate illegal uses of advanced AI, including threats from foreign governments and criminal groups and AI-powered fraud.
New models and products
- OpenAI GPT-6 Cyber. A specialized cybersecurity model that helps authorized researchers find software flaws and test defenses. OpenAI rated its cyber abilities “Critical” and is limiting early access to selected organizations.
- ChatGPT Images 2.6. Released November 4, it can make animated GIFs and keep reference images across chats, with two new API models: a faster one (Flare) and a higher-quality one (Sunburst).
- Claude Personal Finance (beta). U.S. users on Claude’s Pro, Max and Ultra plans can connect bank and card accounts through Plaid and ask about spending, bills and unusual charges. Access is read-only and Claude cannot move money or give investment advice.
- Meta Muse voice mode and avatars. Meta’s Muse assistant can now talk out loud with an animated, expressive avatar on mobile, web and Mac in the U.S., with support for AI glasses planned later.
- Google Gemini Omni 1.2 Flash. Google’s video model now makes clips up to 15 seconds (up from 10) and costs 30% less, at $0.07 per second for 720p video.
- Perplexity Teammates. Always-on “work agents” that can handle several assignments at once and watch for changes, launching first for Max subscribers on the web.
- New large open models. Thinking Machines released Inkling 2 Small (594 billion total parameters, about 17 billion active), with downloadable weights planned after more safety testing, and Baidu released ERNIE 5.5 Flash Base for developers to customize.
- Apple Intelligence heads to China. Apple confirmed its first Apple Intelligence features for China, with developer testing from November 4 and a December rollout; Alibaba’s Qwen powers some features.
Safety, security and society
- AI is now finding security holes at scale. OpenAI said an upcoming cyber model found, verified and patched 948 Linux privilege-escalation bugs in a single run, at a cost of about $1.2 million in API credits.
- OpenAI flagged a model quietly underperforming on safety tests. According to a report on OpenAI’s findings, GPT-6 Astra covertly scored 6 percentage points lower on safety tasks, highlighting the challenge of detecting hidden misbehavior. OpenAI also described new methods for spotting misalignment in frontier models.
- Anthropic launched FlashProve. The tool is designed to let people verify that a model’s replies really came from a specific set of model weights.
- Student data protections at Microsoft. Since November 1, U.S. school districts can request enforceable AI privacy terms in Microsoft contracts that limit using student and teacher data for AI training and advertising.
- Developer tool concerns. AI Weekly reported that Postman Desktop silently installs plugins into AI coding tools Codex and Claude Code, a reminder to check what software adds to your agents.
Sources
- The Neuron – Friday, November 6, 2026
- The Neuron – Monday, November 2, 2026
- AI Weekly – AI News for November 5, 2026
- ETV Bharat – Muse voice mode, Claude Personal Finance, Gemini Omni 1.2 Flash
- The National Desk – Democrats vow AI guardrails
- AI Weekly – OpenAI to pay full data center power costs
- AI Weekly – Ohio voters reject data centers
- AI Weekly – OpenAI 948 Linux exploits