Google turns Gemini into an interlocutor that can think while it speaks

Google presented Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking on September 15, two models designed for real-time voice conversations. The company says the new generation can maintain a more natural dialogue while processing information and carrying out complex tasks. The Extended Thinking version adds an additional layer of reasoning for more difficult problems. Google explains the new technology here.

What is changing: voice is becoming an interface for AI. Instead of typing a command, users can have a conversation in which the system listens, processes information and responds immediately. If this experience becomes natural enough, AI assistants could be used much more often in situations where typing is inconvenient or impossible.

OpenAI discusses new funding at a valuation of around $1.2 trillion

OpenAI is holding preliminary discussions with investors about a new funding round that could value the company at approximately $1.2 trillion, according to the Financial Times. Reuters reported that the negotiations are still ongoing. OpenAI raised approximately $122 billion in March at a valuation of $852 billion. The Financial Times reported on the negotiations.

What this figure shows: developing next-generation AI models has become a business requiring enormous amounts of capital for data centers, processors and energy, not just research. A potential $1.2 trillion valuation would represent a significant increase from March's figure. For now, however, it is a valuation being discussed between the company and investors, not a finalized valuation.

Anthropic shows how far the abusive use of Claude can go

Anthropic has published a new Threat Intelligence report documenting operations across seven areas, including cyberattacks, surveillance, fraud, influence operations, biological activities and weapons development. The company says it identified and blocked the accounts involved and, in some cases, shared information with authorities and other organizations. The full Anthropic report is available here.

Why the report is attracting attention: AI can reduce the time and expertise required for operations that previously needed specialized teams. Anthropic describes nearly autonomous uses in cybersecurity, as well as influence operations in which AI generated large volumes of content and fake profiles. The report also includes the case previously covered by AIdapted involving propaganda in Romania and Moldova, but the new information places that case within a much broader phenomenon.

Meta turns AI from a free feature into a subscription product

Meta has launched Meta One globally, a subscription system offering expanded access to AI features and additional tools for Instagram, Facebook and WhatsApp. The offering includes image and video generation features, creative tools and facilities for creators and businesses. Prices start at $2.99 per month for certain products. Meta presents the new service here.

What it could mean for users: AI is beginning to follow a model already established by other digital services: free access for ordinary use and payment for higher limits or advanced features. Meta is therefore trying to turn AI usage into a direct source of revenue, while taking advantage of its existing billions of users across its platforms.

Threads expands its tools for podcasters

Threads is introducing dedicated podcast profiles, episode links, transcripts, guest tagging and tools for promoting shows. The platform is not becoming a separate audio-hosting service; instead, the focus is on podcast discovery and the conversations that develop around individual episodes. TechCrunch presents the new features here.

Why it matters to creators: an audio episode can more easily become a starting point for written content and online discussions. Automatic transcripts can make episodes easier to search and share, while Threads is trying to occupy the space between a podcast and the community that follows it.

Ursula von der Leyen calls for slowing the development of the most advanced AI models

In her State of the Union address, delivered in Strasbourg on September 16, European Commission President Ursula von der Leyen backed the idea of slowing the development of frontier AI and announced that she intends to invite leading AI laboratories to discussions on model evaluation, early warning systems and security. Reuters reports on the position expressed in the European Parliament.

The dispute is now also political: supporters of a more cautious approach point to the risks posed by increasingly powerful models, particularly in cybersecurity. Industry representatives, including Mark Zuckerberg, take a different view and argue that AI laboratories already have sufficient incentives to manage their own risks. For the EU, the challenge is how to introduce safeguards without slowing European research and development in a market dominated by American and Chinese companies.