93% of children in Romania use a smartphone every day, while AI is increasingly entering schools

According to an EU Kids Online study reported by Euronews Romania, 93% of children in Romania use a smartphone every day. The study covered 28,465 children aged 10 to 16 from 19 European countries. In Romania, the smartphone has become the main gateway to the internet for many children, while 82% said they had used AI at least once during the previous month. Around 45% had used it for essays and other written assignments.

The numbers show that high technology use does not automatically mean strong digital skills. For children, knowing how to ask an AI system for an answer is becoming less important than knowing how to check that answer, understand its limitations and recognize when the system may be wrong.

Romanian teachers are being encouraged to integrate AI into classroom activities

The debate in Romania is increasingly moving from whether AI should be allowed in education to how it should be used. At a conference reported by Radio Romania Actualitati, specialists encouraged teachers to bring AI tools into classroom activities, particularly as students are already using them extensively. Romanian education regulations also increasingly refer to the critical, ethical and responsible use of digital technologies.

This changes more than the teacher's toolkit. It also changes how assignments and assessments need to be designed, because a student can now use AI to produce a text, summarize information or generate ideas in seconds. The teacher's role increasingly includes helping students understand, question and improve what AI produces. The relevant Romanian education legislation already reflects this direction.

OpenAI introduces ad-sponsored agents in ChatGPT

OpenAI has announced that it is testing Sponsored Agents, allowing users to start conversations with AI agents sponsored by companies after interacting with an advertisement. The sponsored interaction is intended to be clearly labeled and separated from independent ChatGPT responses. OpenAI is also developing AI tools that advertisers can use to create, modify and analyze campaigns.

The change could turn an advertisement from a static message into a conversation. Instead of simply seeing a product, users could ask an AI agent questions about it and receive information directly from the sponsoring company. That makes the distinction between an independent AI answer and a sponsored interaction particularly important.

OpenAI creates an official system for reporting “misaligned” AI behavior

OpenAI has published a new Model Misalignment Reporting Framework. The company says reports of potentially concerning model behavior had previously been handled more informally. The new framework is designed to make the reporting and analysis of such incidents more systematic, with examples collected from the previous six months.

As AI systems gain access to more tools and can carry out increasingly complex actions, the question is no longer only whether a model gives an incorrect answer. Researchers and developers also need to understand situations in which a system behaves unexpectedly, attempts to evade oversight or interacts with software in unintended ways. A standardized reporting process makes these incidents easier to study.

AI is beginning to change jobs before it changes their titles

An OpenAI analysis of more than 1.5 million work-related ChatGPT messages from April through July 2026 shows employees using AI for activities traditionally associated with other professions. The analysis also points to repeated use of AI for an expanding range of workplace tasks.

That means changes in the labor market may begin without a job title disappearing. An employee may still officially be a marketer, manager or analyst while spending part of the working day performing tasks that previously belonged to another role. The first visible change may therefore be the content of a job rather than its name.

Researchers are turning scientific papers into AI agents that can “talk” to one another

Researchers at Stanford have developed Paper2Agent, a system designed to transform scientific papers and the methods they describe into interactive AI agents. According to the study published in Nature, these agents can answer questions, apply methods to new data and interact with agents created from other scientific papers.

The idea changes the role of the research paper. Instead of being only something a scientist reads, a paper could become a reusable computational resource. If this approach scales, agents based on different papers could work together, compare methods and analyze data, potentially speeding up parts of the research process.

A US court clarifies an important issue involving AI and copyright

The US Court of Appeals for the Ninth Circuit has rejected an important part of a lawsuit brought by programmers against GitHub, Microsoft and OpenAI over AI coding tools. In the Doe v. GitHub decision, the court addressed claims concerning the alleged reproduction of copyright-management information in AI-generated code.

The decision does not settle every dispute concerning AI training, generated code or open-source licenses. Instead, it clarifies a specific legal issue involving the Digital Millennium Copyright Act while leaving other claims available for further consideration. For developers and AI companies, that distinction matters because copyright questions around AI remain far from resolved.

AI data centers are creating a much larger e-waste problem than previously estimated

An analysis by the Basel Action Network estimates that AI infrastructure could generate between 395 million and 617 million metric tons of discarded electronic equipment between 2025 and 2050. The estimate includes servers and GPUs, but also equipment used for power, cooling and networking.

The environmental cost of AI is therefore not limited to electricity or water consumption. AI hardware is replaced as computing requirements grow and newer chips become available, creating a large physical waste stream. Recycling and responsible disposal could become an infrastructure challenge alongside energy and cooling.

A new industry is emerging around testing and certifying AI agents

AIUC has raised $40 million in a Series A funding round to develop standards for evaluating and certifying enterprise AI agents. According to SecurityWeek, its AIUC-1 framework tests agents against threats including jailbreaks, prompt injection, hallucinations, abnormal behavior and data leakage, using thousands of adversarial scenarios and periodic audits.

The need for certification grows as agents move beyond simple conversations and gain access to email, documents, software systems and business decisions. Companies may increasingly need evidence that an AI agent has been tested before allowing it to operate inside sensitive environments.

OpenAI begins using AI for the advertising industry itself

Alongside Sponsored Agents, OpenAI is developing tools that allow advertisers to use natural-language instructions to create, modify and analyze campaigns. The company's advertising plans also include AI-generated text and images and integrations with services such as HubSpot and Shopify, as described in its advertising announcement.

This creates a broader strategy in which AI can participate on both sides of advertising: helping companies create campaigns while also becoming the interface through which consumers interact with sponsored products. Advertising could therefore become less about displaying a message and more about enabling an AI-mediated product conversation.

Canada and Germany announce investments of CAD 150 million and EUR 100 million in a new approach to AI

Canada and Germany announced planned investments in LawZero, an organization founded by AI researcher Yoshua Bengio. Canada is committing CAD 150 million and Germany EUR 100 million toward research based on a “safe by design” approach to AI, according to the Canadian government's announcement.

The approach differs from the idea of building increasingly capable systems first and adding safeguards later. Instead, safety and control are intended to be part of the system's architecture from the beginning. The investments also show that governments are treating research into AI safety and control as a strategic area rather than only a technical concern for private laboratories.

Sources: