2026-W46 · 14 stories

The week in AI: November 9–15, 2026

The week's main developments, grouped by original publication date. Every item links to our full story and the original source.

Issue date: ·Week: November 9–15, 2026

What changed this week

Three key changes

  1. DeepMind Achieves First Implementation-Level Formal Verification of Signal Code

    What changed
    Google DeepMind researchers used AlphaProof Code to mathematically prove Signal’s compiled protocol code security properties in four weeks.
    Practical impact
    This enables organizations to verify that their compiled cryptographic code implements security guarantees, catching implementation bugs missed during design reviews.
    What to consider
    Requires AI agents running on Gemini Carbon to generate over six million lines of Lean code; applicability to other protocols or custom code without adaptation remains unproven.
    Full story ↗
  2. Gemini Omni 1.2 Flash with SynthID Watermarking and Lower Costs

    What changed
    Google Cloud released Gemini Omni 1.2 Flash with a 30% price cut to $0.07/second for 720p video, extended clip length to 15 seconds, and introduced invisible SynthID watermarks.
    Practical impact
    Businesses can generate branded video content with built-in attribution via SynthID and control audio reference usage through allowlists.
    What to consider
    Allowlists restrict free selection of audio sources, and SynthID watermarks are invisible but may not prevent unauthorized reuse.
    Full story ↗
  3. GitHub Copilot Deprecates Models Starting December 14, 2026

    What changed
    Starting December 14, 2026, GitHub Copilot will deprecate Claude Fable 5, Claude Haiku 4.5, and Gemini 3.8, requiring users to update workflows and integrations.
    Practical impact
    Users must migrate to new models to maintain functionality, especially for Copilot Enterprise and Business customers.
    What to consider
    Individual subscribers with annual plans retain access to Claude Haiku 4.5 but must update workflows and integrations by December 14.
    Full story ↗

Models

1

Google Cloud launches Gemini 4 Hydrogen as short-term GA with 1M-token context

Gemini 4 Hydrogen is a short-term GA release with a unified 1,048,576-token limit, four thinking levels, and a minimum 45-day retirement notice policy.

The model provides a significantly larger context window and new capabilities for agentic workflows, but requires migration from previous versions, including SDK upgrades and removal of deprecated parameters.

Full story ↗Original source ↗

Products

2

Cohere announces completion of AISLE acquisition and welcomes cubic team

Cohere closed its AISLE acquisition and welcomed the cubic team. The company launched Cohere Cyber and a secure coding vertical, but did not provide pricing, metrics, or availability dates.

The acquisitions add cyber defense and code review to Cohere's sovereign AI offering for private cloud, on-premises, and air-gapped deployments; performance claims remain Cohere's own statements.

Full story ↗Original source ↗

Gemini Omni 1.2 Flash: SynthID Watermarking and Enterprise Features on Google Cloud

Google Cloud launches Gemini Omni 1.2 Flash, introducing SynthID watermarking, allowlisted audio references, provisioned throughput, and a 30% price cut to $0.07/second for 720p video.

Business users and developers benefit from a faster, more affordable video generation model with enterprise-grade security features, including watermarking and audio reference controls.

Full story ↗Original source ↗

Applications

3

Microsoft Indonesia Employees Leverage AI to Streamline Workflows

Microsoft Indonesia employees are redesigning recurring workflows with existing tools like Microsoft 365 Copilot and AI agents, for tasks such as assembling weekly forecasts and preparing executive meetings.

This demonstrates AI’s efficacy in boosting productivity and freeing employees’ time for more important tasks such as analysis and decision-making.

Full story ↗Original source ↗

Authentic Brands Group built AuthOS AI system with 50 agents on Google Cloud

Authentic Brands Group built AuthOS on Gemini Enterprise Agent Platform. AI-enhanced Reebok images delivered 60% higher ROAS.

The deployment shows how a major brand management company uses Google Cloud's generative AI stack at scale with attested business outcomes like a 60% ROAS improvement for Reebok images. Vertex AI's modular architecture enables rapid model integration.

Full story ↗Original source ↗

Developers

8

AWS Announces New AI Tools & Models for Builders

AWS expanded capabilities for AI builders with new models like Claude Haiku 5.5 and GPT-6.1 Terra, plus enhancements to AgentCore and cost management.

These updates empower AI builders to choose optimal models, control costs, and build reliable agents, integrating AI into core business workflows with greater ease.

Full story ↗Original source ↗

Google Cloud previews AI SQL functions for AlloyDB

New ai.classify and ai.extract functions allow users to perform text classification and structured data extraction directly within AlloyDB via SQL.

Integrating AI directly into the database reduces operational friction, query latency, and the need to move sensitive data to external model endpoints.

Full story ↗Original source ↗

GitHub Copilot C++ Language Server auto-configures MSBuild builds

The Microsoft C++ Language Server for GitHub Copilot now automatically extracts MSBuild compile commands upon project open. Feature supports only MSBuild projects. Active GitHub Copilot subscription required.

Previously, setting up go-to-definition and find references for MSBuild C++ projects required either a manual compile_commands.json file or an agent to configure. Now, the extractor is bundled and automatically invoked.

Full story ↗Original source ↗

Google Cloud unveils OpenRL, llm-d 1.0, and Agent Substrate at KubeCon 2026

Google Cloud at KubeCon 2026 released OpenRL, llm-d 1.0, and Agent Substrate — an open stack for RL fine-tuning, distributed inference, and safe AI agents on Kubernetes.

Google Cloud consolidates RL post-training, distributed inference, and agent safety in a single open Kubernetes stack, promising higher accelerator utilization and lower latency without vendor lock-in.

Full story ↗Original source ↗

Google Cloud makes Cloud Pathways generally available for distributed ML on TPU fleets

Cloud Pathways is now GA on GKE, bringing Google DeepMind’s single-controller runtime to all customers with elastic training and multi-slice orchestration.

Cloud Pathways gives every Google Cloud customer the same single-controller runtime, colocated Python sidecars, and elastic fault tolerance that power Google’s largest AI workloads, simplifying management of massive accelerator fleets.

Full story ↗Original source ↗