Good morning. Today Google puts cybersecurity at the centre of its newest frontier model, SK Hynix posts a record quarter on AI memory, and robots in Shanghai get a new measure of useful work.

Google unveils Gemini 4 Carbon, a frontier model built for coding, agents and cybersecurity. Announced on Wednesday, just two weeks after Gemini 4 Argon, Carbon tops the Artificial Analysis Intelligence Index v5 at 59, scores 88.2 percent on SEC-bench Pro for finding real software vulnerabilities and 71 percent on CWE-bench for producing working fixes. Early access goes first to selected cyber defenders through the Fairwind program, with voluntary United States government testing before wider release. Why it matters: Google is treating who gets access as part of the launch itself, a clear sign of how powerful it believes this model has become.

SK Hynix posts a record quarter as AI memory demand surges. The South Korean chipmaker reported third quarter revenue of 104.6 trillion won and operating profit of 83.3 trillion won, up 632 percent year on year, as high bandwidth memory for AI systems drove sales. HBM4 is now ramping, long term agreements with 14 major customers cover more than half of planned capacity for 2027 and 2028, and the company says demand is about twice what it can supply, with shortages expected through 2030. Why it matters: the money in super intelligence is flowing first to the memory that makes it run, and that supply is sold out years ahead.

OpenAI is reportedly planning to buy Google AI chips. Reuters, citing The Information, reports that OpenAI is considering purchasing Google Tensor Processing Units, with no price, volume or delivery schedule yet confirmed. OpenAI already rents Google chips and distributes its frontier models through Google Cloud, a deepening tie we followed in yesterday’s edition, and it is also developing custom chips with Broadcom. Why it matters: even the fiercest model rivals are becoming each other’s customers when compute is this scarce.

GitHub says software agents now start a remarkable share of code reviews. In its Octoverse report, GitHub says agents originated more than 40 percent of pull requests in September, up from 17 percent in March, alongside the launch of Copilot Fusion, a model routing system for coding agents, and a new Copilot code review offer. Why it matters: coding is where agents have moved fastest from assistant to colleague, and the share of work they start is now impossible to ignore.

In Shanghai, AGIBOT reframes robots around productive work. At Innovation Day 2026 on Thursday, AGIBOT introduced Worken, a framework that combines the words Work and Token to measure how much useful work a robot performs autonomously and consistently, and launched the G2 Max industrial robot, its GO-3 Preview embodied model and Link-U OS 1.0. Why it matters: for robots and agents alike, the question is shifting from how many are built to how much real work they reliably do.

McMaster researchers release a free tool that speeds the search for new antibiotics. The screening tool predicts which proposed compounds may inhibit harmful bacteria, has already screened more than 12 million molecules and can assess more than 40,000 compounds a day, though every prediction still needs laboratory validation. Why it matters: finding a promising antibiotic candidate has long been slow and expensive, and a free tool that narrows the field this fast could shorten the path to new treatments.

That is the morning edit. We will be back tomorrow with the stories shaping tech and culture.