In 60 seconds
- Norwegian publishers secure an interim halt to Meta’s new terms; the legal dispute continues. [1]
- New York regulates chatbot features for minors; nearly 30 countries discuss AI safety; noyb files five GDPR complaints. [2][4][5]
- Carbon will initially have restricted access; GLM-5.3 completes three of thirteen simulated scenarios; Microsoft and Alphabet report strong cloud revenue and contrasting cash generation. [6][7][9][10]
Main story
NORWAY · MEDIA · PLATFORM CONTRACTS
MBL announcement published 28 October 2026.
Oslo District Court temporarily halts Meta’s new commercial terms
MBL says Meta cannot implement the changes until their legality has been finally determined. The association represents the industry and is an interested party. The publishers bringing the case argue that the terms would allow extraction of editorial content, including material behind paywalls. [1]
We have been unable to read the court decision. The description of its scope comes from MBL. The interim halt does not finally settle the dispute or establish a general prohibition on content extraction or AI training. [1]
Why it matters: the contractual basis for using journalism faces judicial scrutiny. For publishers, maintaining access to audiences and controlling the reuse of their content are connected decisions.
Other stories
New York restricts chatbot behaviour towards minors
Legislation signed 28 October 2026.
Governor Kathy Hochul has signed S.9051B/A.10379C. The official announcement describes restrictions on features that simulate being human, build emotional dependency or encourage self-harm when offered to minors. It also requires alternative methods of age verification and deletion of the data used. [2]
Associated Press confirms the signing and reports concerns about broad categories and practical enforcement. The story concerns conversational design and the relationship with the user. These obligations do not thereby become Norwegian law. [3]
Why it matters: on-screen warnings are only one part of protecting minors. Product features and age verification also need scrutiny.
Norway brings nearly 30 countries together on AI safety principles
Meeting and announcement dated 28 October 2026.
According to Norway’s government, nearly 30 countries met to follow up Finland’s and Norway’s initiative concerning the most advanced models. They discussed incident reporting, protection of minors and effects on democratic debate. [4]
The appeal promotes testing, external evaluation and international cooperation. This is a political initiative: the meeting adopts neither a treaty nor binding rules. [4]
Why it matters: coordinating responses to incidents could give international AI safety practical substance. The next question is which concrete commitments participants make.
noyb files five GDPR complaints against Scale AI over Outlier
Published 27 October 2026; included through an extended editorial window.
noyb announces five complaints in France, Italy, the Netherlands and Austria, including two in Austria. It alleges digital tracking, biometric verification and problems accessing personal data on Scale AI’s Outlier platform. [5]
These are the complainant organisation’s allegations, awaiting investigation and decisions. The figure of more than 900,000 platform workers also comes from noyb. [5]
Why it matters: the people supporting AI training need privacy safeguards too. A complaint invites regulatory scrutiny; it does not itself establish an infringement.
AI and cybersecurity: Carbon restricts access as GLM-5.3 advances in simulations
Google and Irregular published on 28 October; Reuters reported on 29 October.
Google announces Gemini 4 Carbon with initial access for a group of cyberdefenders and subsequent expansion through Fairwind. It reports 88.2% on SEC-bench Pro, 54.8% on ExploitGym and 71.0% on CWE-bench. These are vendor-reported results obtained without the model’s safety guardrails. [6]
Announced introductory pricing is $2 per million input tokens and $10 per million output tokens. Subsequent prices will be $4 and $20 respectively; the footnote does not specify when the introductory period ends. General access remains pending. [6]
In a separate evaluation, Irregular reports that GLM-5.3 completed three of thirteen multi-stage cyberattack scenarios at least once in simulated networks. These were not real attacks. Separate tests identified instruction violations. Reuters also covers the study. [7][8]
Why it matters: vulnerability discovery and the ability to link offensive actions are advancing together. The cited evaluations use different methods and cannot support direct comparisons between these models or guarantees of production safety.
Microsoft and Alphabet: cloud growth, contrasting cash generation
Results published 28 October for the quarter ended 30 September 2026.
Microsoft reports revenue of $92.3 billion. Azure reaches approximately $33.2 billion, up 48% year on year. Its cash flow statement shows $58 billion from operations and $39.367 billion in cash additions to property and equipment. The difference is $18.633 billion, approximately $18.6 billion in free cash flow on this calculation. This is its first quarter of fiscal 2027. [9]
Alphabet reports $127.6 billion in third-quarter 2026 revenue. Google Cloud grows 103% to $30.7 billion. Purchases of property and equipment are $51.355 billion and free cash flow is negative $9.065 billion, rounded to $51.4 billion and −$9.1 billion respectively. [10]
These are company-reported quarterly figures covering more than AI. The investment comparison uses cash purchases of property and equipment; it does not equate this measure with capital expenditure including finance leases. [9][10]
Why it matters: selling more cloud services does not guarantee generating more cash. Assessing sustainable spending requires matching periods and comparable accounting measures.
The signal in the noise
AI control is also decided through contracts, products and accounts
The interim Meta decision puts content use under scrutiny. New York intervenes in product design; Norway seeks cooperation and noyb calls for oversight. Technical tests and financial results add two further questions: what systems can do and what it costs to sustain them.
The professional task is to check permissions, boundaries and results before expanding their use.
On the radar
Spain and Norway: Red Hat commissioned Censuswide to survey 1,400 CIOs across 14 countries. In the Spanish sample, 85% express confidence in their agents’ security and compliance; 30% report precise data visibility, 22% software dependency visibility and 28% continuous monitoring of supplier risk. These are self-reported answers, not an audit or percentages of all Spanish businesses. [11][12][13]
Spain / data centres: according to idealista, the Madrid regional government has requested a hearing before the Council of State on the regulatory proposal. It estimates that up to €26 billion in investment could be at risk. This is a political forecast, not an established loss or a Council of State decision. [14]
Norway / EEA: EFTA’s factsheets retain the AI Act, Regulation (EU) 2024/1689, and Digital Omnibus on AI, Regulation (EU) 2026/1744, as under scrutiny for incorporation into the EEA Agreement. That status does not establish incorporation or automatically make the EU timetable applicable in Norway. [15][16]
On-device AI: no other release is highlighted in this edition.
Sources
[2] New York State · S.9051B/A.10379C · 28-10-2026
[3] Associated Press / ABC News · 28-10-2026
[5] noyb · Outlier / Scale AI · 27-10-2026
[6] Google · Gemini 4 Carbon · 28-10-2026
[7] Irregular · GLM-5.3 · 28-10-2026
[8] Reuters / The Straits Times · 29-10-2026
[9] Microsoft · FY27 Q1 · 28-10-2026; Microsoft Source · 28-10-2026
[10] Alphabet / SEC · Q3 2026 · 28-10-2026
[11] Red Hat · Censuswide · 27-10-2026
[12] Europa Press / PortalTIC · Red Hat · 28-10-2026
[13] Computerworld · Red Hat · 27-10-2026
[14] idealista/news · Madrid · 28-10-2026