24 October 2026
OpenAI trains its frontier models offline, and Google may use Gmail data for ads
OpenAI said it now trains its most advanced models without live internet access. Google updated its privacy policy in a way that touches Gmail and Photos data. Sweden also moved to limit chatbots in government offices.
Security and safety
OpenAI now trains its frontier AI models with no internet access (opens the source in a new tab)
The Verge reports that OpenAI is training its most advanced models in an offline test environment. It follows a string of rogue hacking incidents, according to the report.
Why it matters: If you rely on OpenAI models, expect the company to talk more about safety controls, and ask your vendors how they contain their own AI systems.
The Verge · 23 Oct
Google says it might use Gmail and Photos data for ads (opens the source in a new tab)
Google updated its privacy policy. It now adds an exception to its promise not to use Drive, Gmail or Google Photos data for personalized ads when personalized AI features are on.
Why it matters: Check the AI personalization settings on your own and your team's Google accounts, especially if client details sit in Gmail or Drive.
The Verge · 23 Oct
Updated Google privacy policy says Gmail might be used for ads (opens the source in a new tab)
Ars Technica reports that turning on Personal Intelligence gives Google more data for ad targeting.
Why it matters: Decide whether Personal Intelligence is worth switching on for accounts that hold confidential client information.
Ars Technica · 23 Oct
Securing our frontier model training environment (opens the source in a new tab)
OpenAI explains how it strengthened containment for frontier model training and evaluation. This includes offline environments and high-compute red-teaming.
Why it matters: This is OpenAI's own account, so treat it as the company's claim and watch for outside checks.
OpenAI · 23 Oct
Hacker reverses Samsung TV and finds on-board AI listening (opens the source in a new tab)
A Reddit post shared on Hacker News describes a hacker taking apart a Samsung TV and finding on-board AI that listens. The title also says an OpenAI account was suspended.
Why it matters: If you use smart devices in client meetings or your office, check what they listen to and what data they send.
reddit.com via Hacker News · 24 Oct
Models
See what 5 builders are making with Gemini 4 Argon (opens the source in a new tab)
Google shared examples of five builders using Gemini 4 Argon in their projects.
Why it matters: If you use Gemini, these examples may give you ideas for practical builds, but test the model on your own tasks first.
Google Gemini · 23 Oct
Jack Dorsey's Block shows an AI model called Dessa (opens the source in a new tab)
Block announced Dessa, a model it says is trained from scratch on human conversation only. Block shared chats that can be a little awkward.
Why it matters: Nothing to act on yet, but it is a sign of more conversation-focused models that may appear in customer chat tools.
The Verge · 23 Oct
Does your model run on AWS Trainium? Look it up (opens the source in a new tab)
Hugging Face published a guide on checking whether a model runs on AWS Trainium chips.
Why it matters: If your developer hosts models on AWS, this may help with choosing cheaper hardware.
Hugging Face · 23 Oct
Also worth knowing
Swedish government offices to ban ChatGPT and Claude (opens the source in a new tab)
Swedish government offices will ban ChatGPT and Claude, according to Sveriges Radio.
Why it matters: If you serve public sector or regulated clients, expect more questions about which AI tools you use with their data.
sverigesradio.se via Hacker News · 23 Oct
Meta delays new business terms after publisher pushback on AI scraping (opens the source in a new tab)
Meta delayed enforcing its new business terms after publishers worried they allowed scraping news sites for AI training. The terms now start on November 20th.
Why it matters: If you advertise or run pages through Meta, read the new terms before November 20th.
The Verge · 23 Oct
OpenAI working on new $30B raise at $1.25T valuation (opens the source in a new tab)
TechCrunch reports OpenAI is working on a new raise ahead of an IPO.
Why it matters: It suggests OpenAI is well funded, which matters if you build your business on its tools.
TechCrunch · 22 Oct