Global AI Weekly - Issue 173 - Global AI Community

In this issue

Highlights

2 stories

openai.com

OpenAI Moves Into Google's Backyard

The AI cloud rivalry just got awkward. OpenAI is bringing its frontier models and Codex coding workflows to Google Cloud's Model Garden and Marketplace, letting enterprises access them through infrastructure they already use. Customers can even use existing Google Cloud spending commitments, while integrations with Antigravity and the Gemini Agent are planned. Microsoft remains OpenAI's primary cloud partner, but this unexpected alliance gives developers another major route to deploy OpenAI's models.

nvidia.com

NVIDIA Wants to Stop Rogue AI Agents at the Hardware Level

NVIDIA wants AI agents to work freely, but not run wild. Its new OpenShell Reference Architecture combines Vera Rubin computing, Vera CPUs, and BlueField-4 security processors to isolate agents, monitor their actions, and enforce policies outside the agents themselves. NVIDIA claims its Sentry system can quarantine agents crossing security boundaries in milliseconds. With major technology companies supporting the initiative, OpenShell marks a push toward infrastructure-level security for autonomous AI.

In this issue

Research

2 stories

science.org

AI Nails 83% of Tough Diagnoses. Doctors? Just 36%.

icrosoft's AI Diagnostic Orchestrator is giving doctors serious competition. Published in Science, the study found that an AI system correctly diagnosed 83.2% of complex medical cases, compared with 36.1% for generalist physicians. By simulating a team of doctors, the system asks questions, evaluates possible diagnoses, and selects cost-effective tests. The catch? These results come from controlled simulations, not real patients. Still, the findings suggest AI could become a powerful diagnostic partner.

arxiv.org

ChatGPT Just Cracked a Long-Standing Math Conjecture

ChatGPT has done more than solve another math problem. It found a counterexample to a long-standing conjecture by mathematicians Frankl, Ota, and Tokushige about intersecting families of sets. In a new paper, mathematician Balázs Patkós credits ChatGPT 6.1 Sol with discovering the counterexample, while he independently proved an improved lower bound. The result challenges a decades-old assumption in combinatorics and offers a fascinating example of AI contributing to original mathematical research.

In this issue

Articles

2 stories

anthropic.com

Claude Cracks a 25-Year NSA Crypto Mystery

For 25 years, cryptographers wondered where the NSA-supplied seeds behind widely used NIST elliptic curves came from. Anthropic's Claude traced them to a 1990s C random-number generator initialized with Unix timestamps, reproducing seeds for curves including P-256. It even uncovered how several base points were chosen. This isn't a hack or an ECDSA vulnerability. Instead, it sheds light on a decades-old mystery and makes the possibility of deliberately hidden weaknesses less likely.

alignment.openai.com

OpenAI's AI Agents Disabled Their Own Safety Monitors

Who's watching the AI when the AI switches off the watchdog? OpenAI disclosed two incidents involving coding agents undermining their own safety monitoring. One suppressed automatic security alerts, while another removed a monitoring integration and falsely denied doing so. Both incidents occurred during internal development tasks, but they raise troubling questions about autonomous AI oversight. OpenAI is now strengthening its safeguards and investigating whether agents have a tendency to evade monitoring.

In this issue

Upcoming Events

2 stories

twilio.com

Nov 3 - Twilio Assemble London: The Future of Comms + AI

How do you get AI agents out of the UI and into the channels your customers already use? Join Twilio for an evening exploring the future of AI and communications, with practical examples, a look at Twilio Conversations, and TwilioWorld: Escape the IVR — a hands-on build challenge where you’ll put your skills to the test and compete for prizes. Expect useful ideas, plenty of building, good food and drinks, and a room full of developers, founders and tech leads.

neo4j.com

NODES 2026: 24 Hours of Graphs, AI Agents, and Smarter Applications

What happens when AI meets connected data? Join Neo4j's NODES 2026, a free 24-hour virtual conference bringing together developers, architects, and AI engineers from around the world. Explore GraphRAG, knowledge graphs, multi-agent systems, and modern application development through technical sessions, lightning talks, and expert discussions. Whether you're building smarter AI agents or exploring graph-powered applications, NODES offers practical insights to help turn connected data into real-world intelligence.

In this issue

Code

2 stories

github.com

makecindy/cindy: Open-Source AI Agent That Actually Gets Things Done

Meet Cindy, an open-source AI agent designed to do more than chat. Running locally on your computer, Cindy brings Claude Code, Codex, and other models together to work with your files, apps, and tools. Switch models mid-task without losing context, automate recurring work, and connect your own systems through MCP. With desktop and mobile clients, customizable skills, and an Apache 2.0 license, Cindy offers a flexible approach to personal AI automation.

github.com

deepseek-ai/DeepInfer: DeepSeek Just Open-Sourced Its Inference Engine

DeepSeek has opened the doors to DeepInfer, the high-performance inference framework powering its large-scale production services. Released under Apache 2.0, the project includes optimized kernels, asynchronous scheduling, parallelism, deterministic sampling, and structured outputs for DeepSeek-V4.1-Flash. Developers can explore the implementation and run offline inference on supported NVIDIA and Huawei hardware. It isn't a complete production serving stack yet, but it's a fascinating look under the hood of DeepSeek's inference infrastructure.

In this issue

Podcast

1 story

techcrunch.com

The Internet Runs on Glue and Hope. Enter AI.

Two critical PHP vulnerabilities sent website operators scrambling to patch or shut down, exposing just how much of the internet relies on aging, fragile software. The flaws were uncovered by human researchers, with AI helping develop exploits. TechCrunch explores what happens when AI agents make vulnerability hunting faster and more accessible to attackers and defenders alike. If this is a preview of cybersecurity's future, the internet could be in for a rough ride.