AI News – October 26, 2026: OpenAI on Google, biosecurity and the end of the live web

The AI week (October 19-25, 2026) reshuffles the cards of the market and security. OpenAI breaks the de facto exclusivity with Microsoft by landing on Google Cl

The week between October 19 and 25, 2026 marks a clear transition in the Artificial Intelligence market: competition shifts from pure computing power to enterprise distribution, operational integration, and infrastructural security. Historic alliances are being reshuffled, while governments are beginning to treat AI no longer as simple software, but as a pillar of national security and public health.

Here are the 5 most important news stories of the week, selected and analyzed by the editorial team of La Bussola dell’IA.

1. OpenAI arrives on Google Cloud

On October 21, Google Cloud announced the arrival of OpenAI models (GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna) and Codex agentic workflows within its Model Garden [2546, 2556]. Access is initially in private preview for a select group of enterprises, with general availability expected in the following weeks. Although Microsoft maintains its role as primary cloud partner, this is an earthquake for the enterprise market.

OpenAI ceases to be a de facto exclusive of the Azure ecosystem. Companies will now be able to purchase OpenAI models directly on Google Cloud, unifying billing and leveraging spending commitments already stipulated with Mountain View [2546, 2550].Link: Google Cloud Blog

2. Anthropic launches Claude Opus 5.5 and Claude Duet

On October 25, Anthropic relaunched the challenge by presenting Claude Opus 5.5, a model optimized for agentic coding and knowledge work, with a declared 40% cost reduction compared to the previous Opus 5 on typical workloads [1659, 2530].

At the same time, it debuted in research preview Claude Duet, an agent designed for professionals and SMEs capable of sending morning briefings and operating autonomously on connected applications. The AI race abandons the exclusive metric of text quality to focus on executing multi-step tasks directly on business tools (email, calendars, documents, and management systems) [1659, 2530].Link: Anthropic Newsroom

3. The United Kingdom creates the AI Biosecurity Lab

On October 24, the British government announced the creation of a £100 million AI Biosecurity Lab, the first government organization entirely dedicated to AI-enhanced biosecurity [2543, 2544].

The initiative boasts highly prominent partners: OpenAI and Anthropic are providing up to £200 million in computing credits and model access, while Google DeepMind and Basecamp Research participate as scientific partners. This project marks AI's definitive transition from a productivity tool to a public security infrastructure, aimed at early detection of biological threats and accelerating the development of vaccines and therapies [2544, 2547].Link: GOV.UK — Introducing the AI Biosecurity Lab

4. OpenAI blocks training of frontier models on the live web

In a surprise move on October 23, OpenAI stated that its future frontier models will no longer be trained on real-time copies of the web, relying exclusively on cached versions and simulated web environments [2530, 2537].

The company cites "significant risks" as the basis for the decision. The choice touches three raw nerves in the industry: the security of autonomous agents, the quality and provenance of data, and the increasingly tense and conflictual relationship between AI labs and publishers. While on one hand this closure drastically reduces models' exposure to manipulated content, poisoning, or real-time attacks, on the other it raises serious questions about the freshness of information and the actual representativeness of the data used [2530, 2537].Link: TechCrunch — OpenAI says it will no longer train its AI models on the live internet

5. The EU publishes the first harmonized standard for the AI Act

On October 19, the European Commission published in the Official Journal of the EU the reference to the first harmonized standard in support of the AI Act: the EN 18286:2026 standard [2535].

Dedicated to quality management systems for Artificial Intelligence providers, this standard bridges the gap between legal theory and business practice. After the regulation's entry into force, operational compliance tools finally arrive: for European companies, adopting this standard provides a concrete and certifiable path to demonstrate compliance with the strict requirements set forth by law [2535].Link: AI Policy & Regulation Tracker

Useful Mention: Geopolitics and Procurement

Among the industrial policy news of the week, particular attention is due to the tender issued on October 19 by the United States Department of Homeland Security (DHS), which officially excluded the Chinese company DeepSeek from federal civilian IT systems [2533, 2532]. A decision that confirms how the geopolitics of generative models is now dictating the rules of Western public procurement.

Bibliographic References and Links

Article by the Editorial Team of La Bussola dell’IA