AI: The Most Important News of the Week (September 8-14, 2025)

Apple delays Siri to 2026, Anthropic blocks China, EU sanctions AI. But good news: 99.6% deepfake detection. Complete analysis.

Every Monday, we select and analyze the 5 most significant news stories from the world of artificial intelligence. Not just a simple summary, but a critical reading of the developments that are truly changing the industry. Without hype, without unnecessary technical jargon.

Why 5 stories? Because it's enough to stay updated without being overwhelmed by information.

1. NVIDIA Unveils Rubin CPX: The New Architecture for the Long-Context AI Era

NVIDIA has unveiled its new Rubin CPX GPU architecture, designed specifically to handle AI models with extremely long contexts, up to 100 million tokens. The new chip, arriving by the end of 2026, promises to transform how AI handles video, code, and complex applications.

🔍 What happened: NVIDIA announced the Rubin CPX on September 9th, a new processor that consolidates video decoding, encoding, and AI inference into a single architecture. The company predicts that a $100 million investment in these systems could generate $5 billion in revenue from tokens.

💡 Why it matters: The challenge for modern AI is no longer just speed, but the ability to process enormous contexts. One hour of video requires up to 1 million tokens: current chips struggle. Rubin CPX represents the first attempt to design hardware specifically for this new reality, where AI must "remember" and process massive amounts of information simultaneously. It's the evolution of what we analyzed on the topic of AI and memory: when algorithms remember for us.

🎯 Our take: NVIDIA is betting on a future where AI no longer processes single requests, but manages extended and complex work sessions. It's a vision that anticipates the evolution towards AI agents capable of long-term projects, but the price of these systems risks widening the gap even further between those who can afford advanced AI and those who cannot. As highlighted in our analysis of digital inequality, this could create a further technological divide.

Source: NVIDIA News

2. Anthropic Introduces Memory in Claude and Restricts Access for Chinese Companies

Two significant updates for Claude: on September 11th, Anthropic launched the "memory" feature for Team and Enterprise plans, allowing the AI to remember preferences and projects across different sessions. Simultaneously, on September 5th, the company implemented new restrictions blocking access to entities controlled by Chinese companies.

🔍 What happened: Claude's memory is now active for business subscriptions, eliminating the need to repeat context in every conversation. In parallel, the new policies prohibit access to any organization controlled more than 50% by companies headquartered in China, regardless of their physical location.

💡 Why it matters: Memory represents the evolution of AI assistants from disposable tools to persistent work partners. But the geopolitical restrictions signal a new phase in the "tech cold war": AI is becoming a strategic asset to protect. This could fragment the global artificial intelligence ecosystem. This is what we explored in our article on who controls AI: when technology becomes a geopolitical instrument.

🎯 Our take: Anthropic is balancing innovation and national security, but risks setting dangerous precedents. If every country starts blocking access to its own AI models based on nationality, we could witness the balkanization of global artificial intelligence, slowing progress for everyone. It's a dynamic reminiscent of the information war between national technological systems.

Sources: SiliconANGLE, CRN Asia

3. Apple Delays AI for Siri Again: September 9th Event Disappoints Expectations

The September 9th Apple Event confirmed fears: no AI-powered version for Siri. Despite competitive pressure and market expectations, Apple maintained its focus on hardware, postponing the intelligent evolution of its voice assistant to 2026.

🔍 What happened: The Apple Event presented new iPhones, Apple Watches, and iPads, but no significant announcement about Siri AI. The company confirmed integration with GPT-5 via OpenAI, but without a firm timeline. Analysts now speak of a launch no earlier than 2026.

💡 Why it matters: Apple, once a pioneer in voice assistance, risks falling behind irreversibly. While Google Assistant and Alexa evolve towards natural conversations and Claude introduces memory, Siri remains stuck with predefined commands. In a market where conversational AI is becoming standard, this delay could prove costly for the Apple ecosystem. It's the perfect example of how digital skills become crucial to avoid being left behind.

🎯 Our take: Apple's caution in AI, once seen as wisdom, now appears as strategic paralysis. The company that invented the mobile voice assistant is watching competitors completely redefine the category. The risk is not just technological: it's that users may start perceiving Apple as "the slow one on AI." As analyzed in our piece on failed startups, even giants can fall if they don't adapt.

Source: Mashable

4. Europe Activates AI Act Sanctions: Up to 35 Million Euros for Violations

As of August 2nd, the sanctions of the European Artificial Intelligence Regulation have officially come into force. National authorities can now impose fines of up to 35 million euros or 7% of global turnover for the most serious violations, marking the beginning of the world's first comprehensive AI regulation.

🔍 What happened: The European AI Office became operational on August 2nd, along with the AI Act's sanctioning regime. Fines range from 7.5 million for false information to 35 million for prohibited AI practices. Member States must designate competent national authorities by this date.

💡 Why it matters: Europe has chosen the path of preventive regulation, while the USA and China focus on innovation first and rules later. This approach could protect European citizens from AI abuses but risks slowing technological adoption. The real test will be seeing if the sanctions truly deter misconduct or simply shift innovation elsewhere. As highlighted in our analysis on who decides the rules of the game, regulation is a delicate balance.

🎯 Our take: Europe is writing the first rulebook of the AI era, but with the risk of being the only one to follow it. If the sanctions prove too severe, they could push AI companies to avoid the European market, creating a paradox: protecting citizens by depriving them of the best available technologies. This is the challenge of AI ethics: balancing innovation and protection.

Sources: DLA Piper, European Commission

5. Deepfake Detection Becomes a Priority: Algorithms at 99.6% Accuracy and New Competitions

The fight against deepfakes is intensifying with the launch of new algorithms promising 99.6% accuracy in detection. In parallel, IJCAI 2025 has announced the "Challenge on Deepfake Detection and Localization" with the largest dataset ever created: 1.8 million samples.

🔍 What happened: Attempts at deepfakes against digital platforms now occur every 5 minutes globally, with a 1,300% increase for audio deepfakes and a 700% increase for video deepfakes in 2024. In response, companies like Facia are developing real-time detection algorithms with record accuracy.

💡 Why it matters: 2025 has become the turning point for industrial deepfakes. No longer just simple social media tricks, but sophisticated tools for financial fraud and corporate manipulation. A 2024 case caused $25 million in losses with a deepfake video call of a CFO. Detection is no longer optional: it's digital survival. As we analyzed in our article on artistic deepfakes, the line between art and manipulation is becoming increasingly thin.

🎯 Our take: The battle between deepfake creation and detection is becoming a technological arms race. Every improvement in detection pushes deepfake creators to evolve further. The real winner will be whoever manages to standardize digital identity verification before deepfake technology becomes indistinguishable from reality. It is part of the broader information warfare characterizing our time and is directly linked to the problem of truth in the digital age.

Sources: LinkedIn – Deepfake Detection, IJCAI 2025 Competition

📊 What These Developments Tell Us

This week in artificial intelligence tells us four stories that converge towards an epochal turning point:

Hardware is specializing: NVIDIA Rubin CPX marks the evolution from "general purpose" AI to systems designed for specific tasks. No longer generic chips that do everything, but dedicated architectures to handle the growing complexity of artificial intelligence. It's a signal that AI is maturing beyond the experimental phase.

Geopolitics enters the algorithm: Anthropic's restrictions towards China demonstrate that artificial intelligence is no longer a neutral technology, but a national strategic asset. We are witnessing the birth of "technological blocs" that could fragment the global progress of AI.

Giants slow down (and risk): Apple's delay on Siri AI reveals that even big tech can lose pace in the era of algorithmic acceleration. Caution, once a virtue, becomes competitive paralysis when the market moves at AI speed.

Regulation becomes real: With the activation of the 35 million euro European sanctions, the AI Act moves from theory to practice. Europe is betting on preventive protection, risking slowing innovation to guarantee safety.

The paradox of the week? While we build increasingly sophisticated systems to detect deepfakes at 99.6%, the technology to create them evolves just as rapidly. It's a digital arms race where winner and loser change every week.

The uncomfortable truth: Artificial intelligence is not just changing technology – it is redrawing geopolitical balances, corporate strategies, and the very concept of digital truth. What we are experiencing is not linear progress, but systemic transformation.