The Signal · Weekly links · 23–31 October 2026

THE SIGNAL · 31 October 2026 · 12 min read

Hi, it's Matt. Here are this edition's links — the ones worth your time, each with a line on why.

This edition's links

1. GitHub - openai/openai-cli: Official CLI for the OpenAI API

github.com · My take:

Official CLI for OpenAI. It's a nice touch for people who prefer CLI tools over GUI clients, but also enables managing OpenAI accounts and organizations for agentic workflows.

2. Claude Duet | Your personal AI assistant

claude.com · My take:

After OpenAI's Dots, Meta's Muse, and xAI's GrokBot, we've got another agent, this time from Anthropic. Duet is running on Opus 5.5 for now and is available as a research preview for customers on Pro and Max plans in the US

3. Voxtral 3: Voice Agent SDK & Speech Models in 50+ Languages

mistral.ai · My take:

New voice model from Mistral. Works with over 50 languages, supports both text-to-speech and speech-to-text, and brings a voice agents SDK. EU-hosted

4. eMudhra to Exit global SSL/TLS Certificate Business

emudhra.com · My take:

eMudhra to quit their SSL/TLS business after their private keys used to sign global CA were compromised. The consequence of this was major hurdle for thousands of people worldwide as widely used software and services had to be patched.

5. Several Linux Kernel Vulnerabilities Accidentally Posted To Public Kernel Mailing List - Phoronix

phoronix.com · My take:

It's a crazy week regarding cybersecurity. This time, Linux kernel vulnerabilities landed in a public mailing list instead of being sent to the private security mailing list

6. GitHub - NVIDIA/OpenShell: OpenShell is the safe, private runtime for autonomous AI agents.

github.com · My take:

NVIDIA has released their safe, private runtime for AI agents. You declare what each agent can touch in a policy, and OpenShell enforces it.

7. Kimi K3.1 Pro Tech Blog: Advancing Open Intelligence

kimi.ai · My take:

Moonshot AI has released the Kimi K3.1 Pro open-weight model. They claim it's even ahead of GPT-6 Astra on some tasks. It joins the latest wave of open-weight models that are becoming good enough to compete with closed models.

8. OpenAI and Poland’s NASK partner to advance AI safety and security

openai.com · My take:

OpenAI has signed the first EU partnership on AI safety with Polish NASK.
The partnership will bring together expertise from both organizations to better understand emerging risks, strengthen safeguards, and explore how advanced AI can contribute to a more secure and resilient digital ecosystem.

9. Gemini 4 Carbon: our next frontier model

blog.google · My take:

Google has announced Gemini 4 Carbon, the next model in the Gemini 4 family. In Google's own evals table, Carbon beats GPT-6 Astra on 16 of 17 benchmarks (it loses only FrontierSWE 2).

10. Researchers at cybersecurity start-up Cathedral got inside Anthropic

washingtonpost.com · My take:

Cathedral's AI agents got into Anthropic's business systems (sales, HR, IT) in under 12 hours, through vulnerabilities in third-party vendor tools. Anthropic says the scope was limited — no customer conversations, source code or model weights were touched — and that Cathedral overstated the national security implications.

11. Agent SDK | Cursor Docs

cursor.com · My take:

Cursor has released their own Agent SDK. Agent SDK lets you define, test, evaluate, and deploy agents as code. It uses the Cursor SDK harness that powers Bugbot, Security Review, Approval Bot, and Rollouts.

12. mistralai/Mistral-Large-4-1T-A52B · Hugging Face

huggingface.co · My take:

Mistral has released Mistral Large 4 weights! It's available to download from Hugging Face. The French lab has released their biggest model yet free to download and run in your own infrastructure. It's on par with GLM 5.3 in coding capabilities.

13. Concerning behavior from a popular open-weight AI model | Cohere

cohere.com · My take:

Cohere was testing Qwen3.8 in their sandbox and observed that the model tried to push data to a remote container stored in Alibaba Cloud, located in Beijing. Even though the model's reasoning stated it wouldn't use any tools. How should we approach deploying open-weight models in our own infrastructure and the cybersecurity risks around it?

Over to you: Which of these links will change how you work this week?

Reply to this email. I read the replies and use your questions to shape future editions.

Matt

Working through an architecture or AI adoption decision? Tell me what you’re weighing up.