AI, decoded in real time

Everything AI. Nothing else.

Understand what is changing in AI and what it means for Switzerland and Europe. Five sections covering announcements, practical uses and limitations.

Open the magazine
A Swiss perspective on AI

Sections

Latest

Models & Research

Sam Altman flies to Washington to brief Trump team on GPT‑6 days before cybersecurity deadline

The meeting follows OpenAI's July 21 disclosure of a compromise at Hugging Face, which reporting has linked to the Sol variant. Officials are expected to press for forensic evidence, mitigations and operational…

5 min
Models & Research

Anthropic Rejects Open‑Weights Ban; Amodei Calls Open Models a “Public Good,” Pushes Chip Controls, Distillation Crackdown and Mandatory Tests

Open weights: in this context, “open weights” refers to model parameter files : the numerical tensor values that define a trained neural network : released in a form that allows anyone to run, fine‑tune or redistribute a full model without access restrictions. Publishing weights collapses the compute and expertise required to reproduce model behavior; a party with modest engineering resources can …

21 min
Security & Risks

Autonomous AI Attacks Are the New Normal: Is Your Organization Breach-Ready?

Nvidia launched the Open Secure AI Alliance on July 27 after a July 11–13 autonomous‑AI breach that industry reporting ties to model-driven exploitation, and the move signals defenders must now treat automated, multi‑stage AI attacks as continuous campaigns rather than isolated incidents. The initiative includes roughly 37 founding members and publishes defensive artifacts, including an Apache‑2.0…

10 min
Policy & Ethics

EU’s AI Transparency Guide Arrives as August Enforcement Nears : What Firms Must Do Now

The guide operationalises a set of core transparency duties already present in the AI Act by naming concrete artifacts and data types that will count in supervisory assessments. The practical keywords regulators use are: model documentation (model cards), dataset provenance, training logs and metadata, prompt and chain‑of‑thought records (situational), output labelling and user information, and pe…

18 min
Business & SocietyCH

Chinese Chipmaker Shares Soar Nearly 470% in Blockbuster Debut: What the AI Boom Means for Investors

The stock market reaction that day arrived alongside a broader AI‑driven rotation that hit global chip names and regional indices on 2026-07-28. Reports showed large down moves in established memory and chip makers while investors reallocated into emerging domestic hardware plays, a pattern that framed the debut as a reallocation of capital rather than a standalone phenomenon The Guardian BBC MIT …

9 min
Products & Agents

How Enterprises Must Rebuild Infrastructure to Run Agentic AI at Scale

Enterprises must redesign compute, data, memory, policy and observability stacks to run “agentic AI” : autonomous software agents that execute multi‑step business tasks across people, workflows, data and systems : because agentic systems have fundamentally different production requirements than LLM chatbots, and ignoring those differences will break safety, compliance and reliability at scale MIT …

11 min
Models & Research

Model Context Protocol ships final spec, removes sticky sessions and adds server-rendered Apps

- WorkOS : The biggest MCP spec update ships July 28: What changes for AI agent authentication(https://workos.com/blog/mcp-2026-spec-agent-authentication) - Stacktree : MCP 2026-07-28 spec: what changed, what…

9 min
Security & Risks

China Starts Domestic Immersion DUV Production as Asian Chip Stocks Plunge

- Hardware attestation: Require cryptographic identity and signed firmware for all critical tooling components, and enforce regular remote attestation to detect unauthorized modifications.

4 min
Policy & Ethics

Monday.com Blames AI for Latest Layoffs : A Reverse-Chronological Look at 20 Major Tech Firms

Monday.com is the latest company to link workforce reductions to AI-related strategy shifts, appearing at the head of a reverse‑chronological rollup of similar announcements compiled by TechCrunch on 2026-07-25/26. That running list documents roughly 20 major tech firms this year whose layoff notices, filings or internal communications explicitly referenced AI, automation, product consolidation or…

15 min
Business & SocietyCH

One Fallen Power Line Exposed a Growing AI Data Center Problem : How to Fix It

- Equipment behavior and near miss: Equipment logs reviewed by TechCrunch show repeated automatic bypasses and retransfer attempts that exposed downstream gear to degraded utility conditions. Battery systems and manual intervention prevented a forced shutdown of critical systems, averting immediate data loss or GPU damage, but the incident produced short‑duration brownouts and elevated the risk of…

14 min
Products & Agents

‘Avoiding AI’ Workshops at Public Libraries Surge as Patrons Reject Big Tech

Public libraries have begun running popular “Avoiding AI” workshops that teach patrons concrete, low‑tech ways to limit contact with AI‑driven platforms and data collection, drawing large, diverse audiences and waitlists in many communities, according to TechCrunch TechCrunch.

15 min
Models & Research

Elevated Error Rates on OpenAI Status: Technical Diagnosis and Enterprise Implications

Number of incident entries, operational facts, and root-cause hypotheses recorded

20 min