2026-09-28 to 2026-10-04 digest
11 items on this page: 7 with their own page, 4 short.
A digest puts the short items from one day or one week on a single page. Opened alone, those briefs are often a headline and a sentence, which is a poor fit for a standalone article. Items that meet the importance threshold keep their own pages and are listed here by title.
Items with their own page
Anthropic invests $100 million to train 10,000 engineers via Claude Frontier Academy
Google Project Suncatcher prototype satellite is in orbit
Google introduces Gemini 4 Argon, starting with trusted cyber defenders
OpenAI introduces dots: Astra-powered always-on agents with a stated access boundary
OpenAI introduces GPT-6.1 Sol: an upgrade to GPT-6 Sol, near-Astra on stated tasks, standard API $2 / $10
Anthropic releases Claude Sonnet 5.5: second Claude 5.5 model; faster, lower-cost complement to Opus 5.5
Opus 5.5 has won the timeline, not yet the ledger
Short items
A model guide for the GPT-6 family
【What it is】
On October 2, 2026, OpenAI posted “A model guide for the GPT-6 family.” It tells engineering teams how to pick a model, set reasoning effort and speed, keep prompts, skills, and repository instructions consistent, and move a workflow into production.
The page is in three parts: running in production, adjusting prompts and skills, and keeping long-running work on track. Production covers caching, compaction, measuring success and latency, and monitoring. Long-running work covers steering, async tools, and delegation, including when the model should ask for input. A closing section looks at teams taking GPT-6 Astra from testing to production.
【Verifiable】
Primary: https://openai.com/index/practical-guide-building-gpt-6/
Title: A model guide for the GPT-6 family. Section: Product. Visible date: October 2, 2026. The lede is about choosing a GPT-6 model, giving effective instructions, managing long-running work, and preparing for production.
【Our read】
An engineering guide for the existing GPT-6 family, not a new frontier model, and no new performance numbers others can reproduce.
The latest AI news we announced in September 2026
【What it is】
Google posted a September 2026 AI roundup, dated Oct 02, 2026. It indexes models, products, research, and infrastructure from the month.
The headline is Gemini 4 Argon: complex reasoning, a 1-million-token context window, currently rolling out to trusted cyber defenders through the Fairwind Program. The page also names Gemini 3.8 Flash and 3.8 Flash Cyber, Gemini 3.8 Live and Live Extended Thinking, and 3.8 text-to-speech models.
On products: new Connected Apps in Gemini, Lyria 3.5, Gemini Notebook study tools, CC as a group agent in Google Labs, and Googlebook pre-orders. On research: WeatherNext 3, AlphaGenome Atlas, a methane-tracking model with NASA JPL, a complete adult male fruit-fly brain map, and Project Suncatcher in orbit. It also introduces the DeepMind Institute.
【Verifiable】
Primary: https://blog.google/innovation-and-ai/technology/ai/google-ai-updates-september-2026/
Title: The latest AI news we announced in September 2026. Byline Blog Team. Date Oct 02, 2026. The lede names Gemini 4 Argon, new Connected Apps, and WeatherNext 3.
The Argon section states a 1-million-token context window and a phased Fairwind rollout to trusted cyber defenders before a wider release.
The Suncatcher section describes a test satellite with Planet to see how Google TPUs behave in space. This page is a monthly look-back, not a new launch announcement.
【Our read】
A monthly index, not a new frontier launch. Use it as a September list; the capability claims still belong to the original posts.
OpenAI says it disrupted a coordinated campaign to extract protected model reasoning
【What it is】
On September 30, 2026, OpenAI said it identified and disrupted a coordinated campaign to extract protected reasoning from its models. Earliest activity was in the first week of July. The post says it began July 1 and was fully disrupted by July 28.
【Verifiable】
It calls this adversarial distillation: unauthorized use of one model’s outputs or reasoning to train or copy another. Protected reasoning is the model’s internal working record. Operators, OpenAI says, did not break encryption, breach a database, or read stored chats. They manipulated chats so reasoning became visible, at scale, against the terms—for example by pasting encrypted reasoning into another chat and asking the model to decrypt it. OpenAI says it confirmed cross-model and conversation-compaction paths reported by researchers linked at https://arxiv.org/abs/2608.09867.
July 24–25: 16,000 requests from over 4,000 users. The footnote says these figures are attempts, not necessarily successful extractions. Related prompt patterns spanned a cluster of more than 15,000 users. OpenAI says it is unclear every operator was one actor, and ties a core cluster to individuals associated with Moonshot AI, maker of Kimi.
It reports account bans or limits, stronger signup and monitoring, closing a replay path for encrypted reasoning someone already held, holding streams that might leak reasoning, work with third parties, and sharing via the Frontier Model Forum and government channels. More mitigation is still underway.
【Our read】
This is OpenAI’s account of a terms-of-service campaign it says it stopped, not a broken cipher or direct access to stored chats. Scale on the page is requests and a user cluster; the footnote marks those figures as attempts. Attribution is one core cluster of people tied to Moonshot AI, not every operator. The safety point is OpenAI’s own reason in the post.
DevDay 2026 recap is a short index; the posts that matter are dots and GPT-6.1 Sol
【What it is】
OpenAI published “DevDay 2026 Recap” (https://openai.com/index/devday-2026-recap/). The page is dated September 29, 2026. It calls this the biggest DevDay yet, with more than 20 major announcements.
It also says ChatGPT is opening as a shared surface for people and agents, citing “our collective 1.2B weekly users.”
【Verifiable】
The two items with their own posts are dots (https://openai.com/index/introducing-dots/) and GPT-6.1 Sol (https://openai.com/index/introducing-gpt-6-1-sol/). That is where the substance is.
On this page, dots are always-on agents for Pro and Business Premium in eligible markets. Enterprise, Edu, and Healthcare are an admin-enabled beta, off by default. GPT-6.1 Sol is a major upgrade to GPT-6 Sol: near-Astra ability at a fifth of Astra’s standard input and output token prices, for API, Plus, Pro, Business, Enterprise, and Edu. Dollar prices and eval scores are on the Sol post.
Everything else is one line: Codex and API tooling, ChatGPT plugins and collaboration surfaces, plan and marketplace items, a faster serving tier, and private-deployment previews.
【Our read】
This page is a same-day index. The product detail is on the two official posts above.
The 20-plus announcements are not listed one by one here. The meta description also names GPT-6 Astra. Astra was released on September 3. It is not a new model in this recap.
Aggregated by AGI Pulse. Titles and links only; no full text is republished. Editorial policy