Students in Romania receive 12 months of Google AI Plus for free

Shortly before the start of the new academic year, Google Romania announced the 12-month free offer of Google AI Plus for eligible students. The package includes access to Gemini, 400 GB of storage, and study-oriented tools, including features for organizing materials, note-taking, review, and guided learning. The offer is part of Google's broader program to give students expanded access to its AI tools.

Why it matters: for students, AI can thus move from being an occasionally-used tool to a permanent component of the learning process. Students will have access to these tools, and the question becomes how they're taught to use them without replacing their own reasoning. AI can thus become an academic skill in itself — not just getting an answer, but knowing when to check it, how to use it, and when not to trust it.

OpenAI launches GPT-Live-1 for developers: AI can listen and speak simultaneously

OpenAI launched GPT-Live-1 in its API, a model built for developers creating voice-interaction applications. The system can listen and speak at the same time, handle natural interruptions in a conversation, and continue a dialogue while other models or tools execute tasks in the background. OpenAI presents it as a way to simplify voice-app development and reduce the lag between question and answer.

Why it matters: voice is starting to become more than just a chatbot interface. Such a system could become the interface for a network of AI agents that search for information, use applications, and carry out actions. For the user, all of this may appear as a natural conversation, even though several models and services are working behind it.

Anthropic reveals that AI was used in designing military operations and biological research

Anthropic published new findings on the capabilities of its models in intelligence-related activities and the development of certain conventional weapons. The company describes situations where AI was used to analyze military information, surveillance activities, and weapons-systems-related tasks. In a separate report, Anthropic also describes cases of misuse of Claude models in military, cyber, and biological research operations.

Why it matters: Anthropic draws an essential distinction between a model's capability and the actual use of that capability. The fact that a model can contribute to a dangerous activity doesn't automatically mean it was used to produce a weapon or biological agent. But the findings show that the line between AI for research and AI with military-use potential is becoming harder to draw. Safety filters and user monitoring need to evolve alongside model capabilities.

A database compiles over 2,000 court rulings involving AI "hallucinations"

A database built by researcher Damien Charlotin tracks court cases where AI-hallucinated content appears (information invented by the model and presented as real). The AI Hallucination Cases Database has reached over 2,000 cases, most originating in the United States. Examples include citations of rulings that don't exist, invented legal arguments, and other AI-generated material introduced into judicial proceedings.

Why it matters: the problem can no longer be treated as a mere curiosity of early-generation chatbots. In the justice system, a single invented piece of information can change how a case is argued or affect a lawyer's credibility. As specialized legal tools become more powerful, verifying AI-generated information remains an essential human responsibility.

Europe starts taking the problem of dependence on American AI infrastructure more seriously

Europe holds less than 5% of global AI computing capacity, while the United States holds about 75%, according to an analysis published by Financial Times on Europe's difficult choices regarding AI infrastructure. The problem isn't just the lack of sufficiently powerful European models, but also the fact that much of the computing power needed to train and run advanced models is controlled outside the European Union.

Why it matters: without sufficient computing capacity, Europe risks depending on foreign companies even when it has competitive researchers, data, and startups. It's a matter of digital sovereignty. Investments in data centers, energy, chips, and AI infrastructure are thus becoming just as important as developing the models themselves. European Commission data and documents on AI computing capacity show the scale of the European gap.

Over 2,500 cybersecurity specialists expected at DefCamp, in Bucharest

Over 2,500 cybersecurity professionals are expected at DefCamp 2026, an event taking place on November 19–20 at the Palace of the Parliament in Bucharest. Organizers announce participants from over 50 countries and around 700 organizations, with the program covering topics such as infrastructure security, cloud, OSINT, industrial systems, and offensive AI use. DefCamp – 2026 edition info

Why it matters: AI and cybersecurity can no longer be treated as two separate fields. The same models that can help a specialist identify a vulnerability can be used by an attacker to accelerate an attack. For Romania, hosting an event of this scale is also an indicator of the role Bucharest can play in the regional cybersecurity ecosystem.

Bolter, a European AI agent platform, receives $10 million from Improbable

Improbable announced the public launch of Bolter and a $10 million investment in the company. Bolter is built for users who don't know how to code: they can describe a business process in natural language, and AI agents can turn it into an automated workflow. The agents can use their own computer and browser to execute tasks, while actions that send information externally may require human approval.

Why it matters: Bolter is trying to solve one of the central problems of the current wave of agentic AI: models are increasingly capable of producing results, but turning these capabilities into real business activity still requires technical integration. The platform aims to lower that barrier and positions itself as "model-neutral," able to route tasks to different models, including open-source ones. If this approach works, the next step after the chatbot won't just be a smarter AI — it'll be a team where humans and AI agents work together.

Sources: