
OpenAI hardens GPT-6.2
New protections for internal "thought processes" in Astra and Sol models reduced extraction success rates from 54% to 0.2%.
ddNews Β· a daily AI briefing for Dirk (generative video/ddStudio, Lit

New protections for internal "thought processes" in Astra and Sol models reduced extraction success rates from 54% to 0.2%.
As GPU costs climb, the AI data layer is becoming the primary deciding factor for enterprise investment success.

The app can now autonomously navigate desktop programs, local files, and web pages.

A new GPU compiler extension to better exploit modern hardware architectures.
The need for control over agentic workflows is reviving the "own vs. rent" compute debate.

A confidence-score system that routes easy queries to small models and complex ones to larger LLMs.

Following the resignation of Jensen Huang, the company reports healthy growth despite the CEO scandal.

A Go-based redesign reduced memory footprint by 90% while maintaining sub-1ms latency.

Proposed taxes for companies moving AI data center infrastructure outside the US.
Infrastructure limits in power and thermal management are now more critical than GPU availability.

Google donated its agent infrastructure components to the CNCF for vendor-neutral governance.

An Anthropic mathematician claims to have proved the Riemann hypothesis using Claude.

Nvidia may acquire the diffusion-based LLM startup for over $500 million.

The open-source agent from Nous Research is now available for enterprise integration within Microsoft Teams.

The app is restricting screenshots to prevent OS-level AI agents from accessing private messages.

A free, CC0-licensed local emulator for 200+ services, written and maintained entirely by AI agents.
US lawmakers are creating a new group to tackle security risks of sophisticated AI.
Funding for AI-driven industrial monitoring to detect equipment malfunctions.

A new approach for estimating dense metric depth in images and video with improved temporal consistency.
The app can now interact with local files, desktop apps, and the web to perform multi-step tasks.
A specialized model designed to identify root causes of production alerts and act as an on-call AI agent.
AI models focused on accelerating small molecule research for pharmaceuticals.
Enterprise AI is creating new requirements for non-GPU infrastructure as applications move to production.

The CEO's exit follows a scandal, leaving Colette Kress as interim CEO during a critical US-China infrastructure race.
Despite the leadership change, revenue more than doubled year-over-year with the Vera Rubin platform.
A shift toward power-efficient chips allows enterprises to run AI on existing on-prem hardware rather than costly GPU racks.
Uses stacked 3D DRAM to solve memory bandwidth bottlenecks specifically for AI inference workloads.

The Ramp-AI Index shows a cooling of pricing dynamics despite record-high usage volumes.
Emphasis on open table formats like Apache Iceberg to bridge the gap from AI pilot to production.

Job listings reveal Apple is building an agent capable of coding, web browsing, and cross-app task execution.

A massive sandbox environment used to train Chrome AI agents without risking live web instability.

Shares opened at $111, significantly above the initial $83 IPO price.
The Chinese firm is expanding automated life sciences infrastructure for scientific discovery.

Tightened export controls led to the cancellation of a $150M deal and delivery halts from TSMC and Samsung to China.

The company debuted under ticker "ANT" at $111.50, reaching a valuation of approximately $1.5 trillion.
New hardware trends focus on memory pooling to allow racks to resize based on the specific requirements of different models.

The app can now execute multi-step tasks across local files, desktop applications, and the web.
Enterprises are bolting on cooling infrastructure to adapt legacy data centers for high-density AI racks.
A shared workspace for human engineers and AI coding agents to collaborate on observability.
The proactive multi-agent platform is now integrated across the entire Superhuman suite.

The central agent is being integrated directly into Docs and Mail.
A new startup focused on securing AI agents across cloud infrastructure and SaaS endpoints.
Industry shift toward profitable, durable on-prem "AI factories" to avoid vendor lock-in.
Concerns grow that AI success leads to inescapable vendor lock-in for revenue-generating models.

Introduces ASP.NET Core union type support and new Blazor AI components.

A new global rollout allows real-time transcription as the user speaks rather than waiting for the recording to end.

A new frontier model for software development and science, significantly faster and cheaper than its predecessor Carbon.

Moving away from Electron/React Native to use Zed's GPU-accelerated rendering for better performance.
Google sets costs at $3 per 1M input and $15 per 1M output tokens to compete with GPT-6.2 Sol and Claude Opus 5.5.

Microsoft's Satya Nadella expressed confidence in Kress following Huang's departure.

Microsoft's CEO predicts AI agents will disrupt advertising-driven web business models.

A world model allowing users to create and modify interactive 3D environments in real-time via text prompts.

Using proprietary models, OpenAI achieved a 14% PPA improvement for AMD's flagship GPU-IP in under five months.

Reduced daily shuffle fetch failures from 750 to 20.

A TypeScript library allowing AI agents to compose accessible UIs from reusable behaviors.

New benchmarks for November 2026 compare pricing for Veo, Kling, Runway, and Seedance across various video lengths.