US calls on G20 to avoid new AI rules

The United States has called on G20 countries to avoid introducing new rules that could slow the development of artificial intelligence. At the G20 technology meeting in North Carolina, Michael Kratsios, the White House technology adviser, presented the “Carolina Principles,” a set of principles emphasizing investment in fundamental research and limited government intervention. Washington also opposes the creation of new international AI oversight bodies. According to Reuters, the US position comes at a time when Washington is seeking to maintain its technological advantage over China and prevent regulation from slowing the development of American companies.

Why it matters: two clearly different global approaches are emerging. The US is emphasizing innovation and competitiveness, while the European Union has already built a system based on rules, obligations and risk assessment. AIdapted has already followed the evolution of European AI Act rules. The difference between the two models could influence where AI products are developed in the coming years, the cost of compliance and how much freedom companies have to introduce new technologies.

India prepares autonomous payments by AI agents through UPI

India is preparing a system that would allow AI agents to make certain digital payments through the Unified Payments Interface (UPI), without users having to approve each transaction separately. Initial applications are expected to focus on small, repetitive payments, while the framework includes spending limits, identity verification, predefined rules and mechanisms for delegating funds. According to Reuters, the system is expected to be presented at the Global Fintech Fest in Mumbai. UPI processed approximately 24.51 billion transactions in August 2026, worth around $314 billion.

Why it matters: this is an important step from AI that merely recommends an action to AI that actually carries it out. Within limits set by the user, an agent could purchase a product, pay for a service or manage a transaction without requiring confirmation every time. At this point, the main issue becomes control: what the agent is allowed to do, what limits it has, how it can be stopped and who is responsible if it makes a wrong decision. It is one of the clearest examples yet of the transition toward autonomous AI agents.

Helsinki wants to use heat from an AI data center to heat the city

Finnish company OnZero and energy provider Helen have signed an agreement under which waste heat produced by a new AI data center in Helsinki will be fed into the city's district heating network. When operating at full capacity, the center could supply more than 500,000 MWh of heat per year, equivalent to the heating needs of approximately 70,000 apartments. Heat recovery is scheduled to begin in 2027. According to Helen, the project expands the use of heat generated by data centers in the Finnish capital's heating system.

Why it matters: AI data centers consume enormous amounts of electricity and simultaneously generate large quantities of heat that normally have to be removed through cooling systems. Helsinki is trying to turn this problem into an energy resource. AIdapted has already examined the pressure that AI development is placing on Europe's energy infrastructure in its article on the limits of power grids for data centers. Heat recovery does not solve the electricity consumption problem, but it can reduce part of the energy losses associated with AI infrastructure.

Romania: OVES Enterprise integrates Nemesis AI into STARK autonomous systems

Romanian company OVES Enterprise has entered into a partnership with German military drone manufacturer STARK to integrate Nemesis AI technology into VANTA autonomous maritime vessels. Nemesis AI is designed to enable autonomous systems to interpret sensor data, avoid obstacles and navigate complex environments, including situations where GNSS (satellite positioning system) signals are unavailable. According to SeeNews, the partnership also aims to explore the integration of VANTA systems into Romanian defense programs for monitoring and protecting the Black Sea.

Why it matters: this is one of the developments showing Romania moving from using autonomous technology to developing AI components for such systems. For the Black Sea, an autonomous vessel capable of navigating, interpreting its environment and reacting without constant human intervention could have surveillance and protection applications. The subject naturally connects with AIdapted's analysis of Ukraine and the acceleration of autonomous systems on the battlefield: the experience of war is rapidly turning AI autonomy from an experimental technology into a military capability pursued by states and companies.

Brazil seeks to control AI ahead of the elections

Brazil is trying to limit the impact of AI on the presidential election scheduled for October 4, while the country's Superior Electoral Court is expected to establish new limits on the use of AI-generated or AI-modified content in election campaigns. Rules already introduced include mandatory labeling of AI-generated content, a ban on deepfakes and restrictions on generative systems that recommend or rank politicians. The issue became more urgent after the party of Flávio Bolsonaro presented an AI avatar of former president Jair Bolsonaro. According to Reuters, the Superior Electoral Court had already received 19 cases related to the use of AI during the first 11 days of the campaign.

Why it matters: Brazil is becoming one of the most important electoral tests for the relationship between AI and democracy. The problem is not simply the existence of fake images or videos, but the speed at which they can be produced and distributed to millions of voters. Authorities must determine where legitimate political content ends and manipulation begins. Brazil's experience could become a reference point for other countries holding elections in an environment where AI can almost instantly fabricate statements, images or videos.

WHO warns that AI in medicine should be assessed by institutions' ability to control it

The World Health Organization warns that progress in AI adoption across healthcare systems should not be measured by how quickly the technology is introduced, but by institutions' ability to control it responsibly. The report published by WHO's Regional Office for Europe is based on an international community of experts from 105 countries and identifies fragmented or biased data, unclear responsibilities and insufficient AI literacy as major obstacles. According to WHO Europe, institutions need to develop governance rules, validation mechanisms and the ability of healthcare staff to understand the technology simultaneously.

Why it matters: in medicine, an AI system is not simply a software tool. Its recommendation can influence a diagnosis, treatment or decision about a patient's care. Technical performance alone is therefore not enough: the institution using the system must be able to verify its results, identify errors and intervene when the AI cannot be trusted. WHO's message is that a high-performing AI system is not automatically an AI system suitable for a hospital.

Physical Superintelligence launches an AI laboratory for designing physical systems

Physical Superintelligence (PSI), a startup based in Cambridge, Massachusetts, has emerged from stealth (the period during which a company develops its technology without publicly revealing it) with a $58 million seed round led by Breakthrough Energy Ventures. The company plans to build an AI-based physics research laboratory in which “virtual physicists” can break problems down into testable hypotheses, test them in parallel and build more accurate models of the physical world. Information about the launch and funding was reported by Dealroom.

Why it matters: PSI is attempting to take AI beyond analyzing existing information and use it as a tool for scientific discovery and engineering. The difference is significant: a conventional AI model can suggest a solution, while a system of this kind attempts to formulate hypotheses, test them and repeat the process. If the approach works at scale, AI could become a tool for designing materials, energy systems, sensors and other physical technologies, rather than being limited mainly to software.

Germany: INLEAP Photonics raises €20 million for anti-drone technology

German deep-tech company INLEAP Photonics has raised €20 million in a seed funding round led by UVC Partners. The Hanover-based startup develops high-precision laser systems capable of neutralizing drones and will use the funding for research, production, testing and commercial expansion. According to EU-Startups, the technology is intended, among other applications, to protect critical infrastructure.

Why it matters: inexpensive drones can force defenders to use much more expensive ammunition to intercept them, creating an economic problem that is almost as important as the military one. Laser systems attempt to change this balance by using directed energy to neutralize a target without consuming a missile for every drone. The development of these technologies shows how closely sensors, AI, automation and directed energy are now being combined in the next generation of European defense systems.

Germany: Cloover launches an AI platform for residential energy infrastructure

German company Cloover, based in Berlin, has announced the launch of a platform it describes as an “AI-native neo-utility,” combining financing and installation of solar panels, batteries, heat pumps and other residential electrification systems. The company says it has reached an annualized revenue run rate of more than $350 million and secured a new $100 million financing facility, bringing its total financing capacity to more than $1.3 billion. According to Cloover, the platform can coordinate the production and consumption of energy across homes through a virtual network.

Why it matters: Cloover is attempting to turn homes from simple energy consumers into active components of the electricity grid. Its algorithms can estimate each home's production and consumption and, within limits set by the homeowner, coordinate battery use or heat-pump operation according to energy prices and grid needs. If the model expands, thousands or millions of homes could operate together as a “virtual power plant,” without each household needing its own power station.

Sources