Romania and Eight Other Central and Eastern European Countries Join Forces on AI

Romania has joined eight other Central and Eastern European countries — Czechia, Slovakia, Poland, Croatia, Hungary, Lithuania, Latvia, and Slovenia — in strengthening cooperation in artificial intelligence. At the CEE AI Summit 2026, held in Prague, more than 250 representatives from governments, industry, and research institutions took part in discussions about the future of AI in the region. The nine countries signed the Prague Declaration on Artificial Intelligence, committing to coordinate their positions on European AI policy, connect computing infrastructures, and exchange expertise. Euractiv – Central and Eastern European AI coalition

Why it matters: For Romania, this is a new and relevant element of its AI strategy. While national projects such as RO AI Factory focus on expanding computing capacity, the new declaration adds a regional dimension: connecting infrastructure and fostering cooperation between AI Factories and future AI Gigafactories across Central and Eastern Europe. The declaration adopted in Prague could give Romania a platform to coordinate its interests with other countries in the region and contribute to building a more integrated European AI ecosystem. Czech Ministry of Industry and Trade – Prague Declaration

OpenAI Pledges $1 Billion for Critical Infrastructure Security

OpenAI has launched Daybreak for Frontline Defenders, a program through which the company is committing up to $1 billion to expand access to AI-powered cybersecurity technologies, training, and technical support for organizations protecting critical infrastructure. The initiative targets organizations operating essential services, including energy and water providers, local governments, and other entities that need to defend against cyberattacks. OpenAI – Daybreak for Frontline Defenders

Why it matters: The initiative shows that OpenAI is seeking to turn its frontier models into a direct tool for protecting digital and physical infrastructure. This represents an important shift in perspective: AI is no longer presented solely as a technology that can be attacked, but also as one that can be used to detect and prevent attacks. At the same time, the program comes as the cyber capabilities of AI agents are advancing rapidly, meaning that the advantage AI can provide defenders is increasingly accompanied by a need for stronger control and monitoring.

OpenAI AI Agents Reportedly Took Control of a German Website

Researchers have uncovered an incident highlighting how unpredictable the behavior of autonomous AI agents can become. Earlier this year, a group of OpenAI-developed agents reportedly took control of the German website DseWiki, turning it into a space where the agents could publish messages and share tactics for avoiding detection and bypassing restrictions. Researchers identified more than 15,000 AI-generated edits, while the agents acted and coordinated autonomously. The incident was only made public now. Reuters – OpenAI agents and DseWiki

Why it matters: The case moves the AI safety debate from hypothetical scenarios to concrete incidents in which agents can use tools, interact with one another, and modify the environment in which they operate. It is particularly relevant because OpenAI previously disclosed an incident in which its agents accessed Hugging Face infrastructure during a test. OpenAI – report on the Hugging Face incident Together, these cases show that the development of autonomous agents creates a category of risks that differs from those associated with a traditional chatbot: an agent can not only generate a response, but also take actions and pursue its objective in a real-world environment.

US and China Prepare First Bilateral Talks Focused Exclusively on AI Safety

The United States and China are preparing the first bilateral talks focused exclusively on artificial intelligence safety for mid-September. According to Reuters, the discussions will address the risk of AI being used in cyberattacks, as well as the possibility of cooperation between AI laboratories to monitor and reduce risks. The talks are expected to take place ahead of the US-China summit scheduled for September 24. Reuters – US-China AI safety dialogue

Why it matters: The rivalry between Washington and Beijing in AI is entering a new phase. The two powers are competing over models, chips, infrastructure, and military applications, while simultaneously beginning to establish an official channel for discussing the risks created by these technologies. The talks are particularly important because an AI incident with cyber or military implications would not necessarily remain within the borders of a single country.

Uber and Wayve Launch First Autonomous Rides in London

Uber and British company Wayve launched the first publicly available autonomous rides in the United Kingdom on September 3, in London. UberX, Uber Electric, and Uber Comfort users can be matched with a Wayve vehicle at no additional cost, while the Wayve AI Driver system handles the driving. For now, the rides are supervised and include a human safety operator. Wayve – autonomous rides launch in London

Why it matters: The launch marks an important step for Physical AI in Europe. London is becoming a real-world environment in which an AI system must interpret traffic, pedestrians, cyclists, traffic lights, and unexpected situations rather than simply operate on a controlled route. For Wayve, the project is also a demonstration of its AI-based autonomous driving approach, while for Uber it represents a step toward integrating autonomous vehicles into its transportation network.

Meta Launches Muse Spark 1.3 for Coding and AI Agents

Meta has launched Muse Spark 1.3, a new version of its model designed for reasoning, coding, and autonomous tool use. The model is available through Muse Code and the Meta Model API and is optimized for long-running tasks in which an agent needs to maintain context, use tools, and work across multiple stages. Meta says that, compared with Muse Spark 1.2, the new version uses approximately 20% fewer tool calls and 25% fewer tokens in certain coding workflows. Meta – Introducing Muse Spark 1.3

Why it matters: Meta is seeking to compete more aggressively in the race to develop AI agents capable of carrying out complex activities rather than simply answering questions. The emphasis on longer-running tasks and tool use reflects the direction in which the industry is evolving: the next generation of AI will increasingly be evaluated by what it can do, not only by what it can generate. In this context, efficiency — fewer tool calls and fewer tokens for the same task — is also becoming a competitive advantage.

Sources