Google AI News: Gemini 3, Pentagon Contract, and the Race Against OpenAI

Last updated: December 14, 2025

Google's AI push has accelerated dramatically in the final months of 2025. From Gemini 3's launch to a Pentagon contract putting AI on every military desktop, here's what you need to know, and what it means for your privacy.

The Big Picture

Three things stand out from the past two months:

Gemini 3 is Google's most capable model yet, and it sparked a "code red" at OpenAI
Google's AI now reaches 2 billion users monthly through Search alone
The Pentagon just gave 3 million personnel access to Gemini: the largest government AI deployment ever

November 2025: Gemini 3 Launch

Google launched Gemini 3 in mid-November, calling it their "most intelligent model" to date. Unlike previous releases where Google waited before making new models default, Gemini 3 rolled out immediately across all Google products: the Gemini app, AI Mode in Search, AI Studio, and Vertex AI.

What's New in Gemini 3

Better at understanding context: Google says it "figures out the context and intent behind your request" with fewer clarifying prompts needed
PhD-level reasoning: Gemini 3 Pro scored 1501 on LMArena and jumped to the top of AI leaderboards
"Most factual" model: Trained to minimize hallucinations, though testing by Vectara still found a 13.6% hallucination rate
Nano Banana Pro (officially "Gemini 3 Pro Image"): New flagship for image generation and editing

NotebookLM Gets Smarter

Google's NotebookLM app gained flashcard and quiz generation for studying, plus a Deep Research mode that plans and executes multi-step web research and returns source-grounded reports.

Gemini Live Updates

Gemini Live now supports adjustable speech speeds and more natural two-way conversations. Google Maps and Android Auto got Gemini integration for hands-free conversational assistance while driving.

$40 Billion Texas Investment

CEO Sundar Pichai and Texas Governor Greg Abbott announced a $40 billion Google investment in Texas for AI and cloud infrastructure.

December 2025: The OpenAI Fight Heats Up

Gemini 3's strong performance reportedly triggered a "code red" at OpenAI. According to Axios, CEO Sam Altman paused non-core projects and redirected teams to accelerate development.

December 11: Same-Day Showdown

On December 11, both companies launched major updates within hours of each other:

Google: Released Gemini Deep Research, a "reimagined" research agent based on Gemini 3 Pro
OpenAI: Launched GPT-5.2 (codenamed "Garlic")

The benchmarks are mixed. GPT-5.2 beat Gemini 3 Pro on software engineering tasks (55.6% vs 43.3% on SWE-Bench Pro) and abstract reasoning (86.2% vs 75% on ARC-AGI-1). But Gemini Deep Research topped both on Google's own research benchmarks and Humanity's Last Exam.

Altman downplayed the impact in a CNBC interview: "Gemini 3 has had less of an impact on our metrics than we feared."

December 9: Google AI Goes to War

The Pentagon announced that Google's Gemini for Government will power GenAI.mil, a new AI platform rolling out to 3 million Defense Department personnel.

The $200 Million Contract

Google Public Sector won a $200 million-ceiling contract with the Chief Digital and Artificial Intelligence Office (CDAO). This follows similar $200 million contracts awarded to Anthropic, OpenAI, and xAI in July 2025 to develop "agentic AI workflows" for military applications.

What the Pentagon Is Using It For

Personnel onboarding
Automating administrative tasks
Accelerating contract workflows
"Agentic AI workflows across a variety of mission areas"

Defense Secretary Pete Hegseth: "We are pushing all of our chips in on artificial intelligence as a fighting force."

Undersecretary Emil Michael: "There is no prize for second place in the global race for AI dominance."

Security Details

Deployed at IL5 (Impact Level 5) security classification
Processed in "software-defined sovereign cloud environment"
Google claims Pentagon data "is never used to train Google's public models"
Currently for "unclassified work" only

AI Overviews: 2 Billion Monthly Users

Google disclosed that AI Overviews (the AI summaries that appear at the top of search results) now reach 2 billion monthly users across 200 countries and 40 languages. That's up from 1.5 billion in May 2025.

The Impact on Publishers

This is bad news for websites:

When AI Overviews appear, only 8% of searches result in a click, down from 15% without Overviews (Pew Research)
Clicks on source links within AI Overviews happen just 1% of the time
Ahrefs found a 34.5% lower click-through rate for top-ranked results when AI Overviews appear

Google is now testing a merger of AI Overviews with AI Mode, which could push AI-generated content even more prominently into search results.

Gemini App Growth

The standalone Gemini app now has 650 million monthly users, up from 450 million in July. AI Mode has over 100 million users in the US and India.

Coming in 2026: Google AI Glasses

On December 8, Google confirmed it's developing two types of AI-powered smart glasses for 2026 launch:

Audio-Only Glasses

Built-in speakers, microphones, and cameras
Speak to Gemini for real-time help
Take photos and ask questions about surroundings
No display

Display Glasses

Same AI features plus in-lens display
Turn-by-turn directions
Live translation captions
Notification overlays

Both will run Android XR and connect to a smartphone for processing. Hardware partners include Samsung, Warby Parker (Google committed $75 million to support development), and Gentle Monster.

Project Astra Integration

The glasses will incorporate Project Astra, Google DeepMind's "universal AI assistant" that streams one frame per second from the camera into an AI model for real-time processing. It maintains a 10-minute memory of surroundings and conversations.

Translation: Google glasses will continuously capture and analyze your environment, storing context about where you are and what you're doing. This is "always-on" AI surveillance as a consumer product, the same concern raised over Meta's Name Tag smart glasses.

Privacy Implications

Here's what this Google AI expansion means for privacy:

Data Collection Escalation

As TechCrunch noted in December: "Unlike opt-in services, avoiding Google's data collection may become harder as AI becomes central to its products."

Google's Gemini already accesses:

Emails (Gmail)
Documents (Docs, Drive)
Photos (Google Photos)
Location history
Browsing behavior (Chrome)
Voice recordings (Assistant)
Search history

AI glasses will add continuous visual capture of your environment.

Human Review of Conversations

Google's Gemini privacy policy warns that "human reviewers may read" your data. Even after you delete Gemini activity, conversations reviewed by humans are retained for up to three years.

The Pentagon Question

Google claims military data won't train public models. But 3 million government employees now use Gemini daily, creating massive amounts of interaction data. The question isn't just whether Google trains on this data. It's what access Google (and by extension, law enforcement) has to government personnel's AI conversations.

What Stanford Researchers Say

A Stanford study found that six leading AI companies feed user inputs back into their models. When asked if users should be worried about their privacy, researcher Jennifer King answered: "Absolutely yes."

An Incogni report named Google's Gemini among the "most aggressive in data collection, often lacking transparency."

How to Protect Yourself

Limit Gemini Data Collection

Go to myactivity.google.com/activitycontrols
Turn off "Gemini Apps Activity"
Note: Even with activity paused, conversations are retained for 72 hours

Opt Out of AI Training

If "Keep Activity" is off and you don't submit feedback, Google claims not to use future chats for training
For Workspace users: Enterprise accounts have different (usually better) data protections

Avoid Sharing Sensitive Information

Google AI privacy expert Harsh Varshney recommends never sharing with AI chatbots:

Social Security numbers
Credit card information
Home addresses
Medical records
Confidential work information

Consider Alternatives

For privacy-focused AI options, see our guides on:

Private Search Engines. Avoid AI Overviews entirely
Running AI Locally. Keep your data on your own hardware

What's Coming in Early 2026

AI glasses launch: Google's first consumer wearable AI hardware
Gemini 3 expansion: Expect more integrations across Google products
AI Mode/Overviews merger: Testing now, likely broader rollout
Deeper military integration: The Pentagon's AI push is just beginning
Continued OpenAI competition: The benchmark war will intensify

References

This article will be updated as new developments emerge. Check back for the latest Google AI news.