Latest events in the world of AI — translated and adapted automatically 30 articles
Anthropic CEO Dario Amodei announced plans to legally challenge the U.S. Department of Defense's decision to classify his AI company, developer of Claude, as a supply chain risk. Amodei claims the designation is unfounded and will affect only a minor portion of clients, but the legal challenge could set a significant precedent for the entire AI industry.
Startup DiligenceSquared is deploying voice AI agents to automate a key due diligence step—interviewing a target company's customers. The technology aims to replace expensive consultants, making pre-merger and acquisition checks faster and significantly cheaper for investment firms.
Amazon Web Services (AWS) has announced Amazon Connect Health, a specialized AI platform designed to automate routine tasks in healthcare facilities, such as patient scheduling and documentation, to reduce staff workload.
The US administration is developing radical export control rules that would allow tracking and regulating any chip transaction worldwide, regardless of the country of origin. This unprecedented expansion of restrictions could reshape the entire global semiconductor supply chain.
The U.S. Department of Defense has officially designated AI company Anthropic, creator of Claude, as a supply chain risk—the first such case for an American firm. Despite this classification, the Pentagon continues to utilize Anthropic's AI for operational purposes, including data analysis related to Iran.
The U.S. Department of Defense has officially designated AI company Anthropic, creator of Claude, as a 'supply chain risk.' This unprecedented move against a domestic company stems from Anthropic's refusal to weaken its usage policy, which prohibits applying its AI to autonomous weapons and mass surveillance. The conflict threatens to escalate into litigation.
Netflix has announced the acquisition of InterPositive, an AI startup specializing in film production tools founded by actor and director Ben Affleck. Affleck will also join the streaming giant as a senior advisor, with the deal aimed at integrating unique AI technologies into Netflix's production pipeline.
OpenAI has released GPT-5.4, a new model with native capabilities to control a user's computer, marking a key step toward autonomous AI agents. The model also shows improvements in reasoning, factual accuracy, and code handling.
Meta contractors in Kenya reviewed confidential videos captured by the company's smart glasses, including bathroom visits and intimate moments. A Swedish media investigation has led to a class-action lawsuit against Meta, accusing the company of privacy violations and false advertising.
Apple Music is launching a 'Transparency Tags' system for voluntarily labeling content created using artificial intelligence. The new metadata system covers tracks, compositions, covers, and music videos but currently relies entirely on the honesty of labels and distributors.
Donald Trump has secured promises from major data center operators to independently fund the construction of new power generation capacity to cover their growing energy demands. However, experts doubt the effectiveness of this informal agreement, citing the lack of legal enforcement mechanisms and the questionable economics of such projects.
IT consulting giant Accenture has acquired Ookla, the company behind Speedtest and Downdetector, from media holding Ziff Davis for $1.2 billion in cash. The deal aims to integrate network performance data into Accenture's services to optimize 5G, Wi-Fi, and, crucially, artificial intelligence infrastructure.
New research reveals a concerning ability of large language models (LLMs) to de-anonymize users. By analyzing thousands of posts, AI can match anonymous texts to their real authors with high accuracy, threatening online privacy and challenging the very concept of pseudonymity in the digital age.
Google has unveiled a plan to protect HTTPS certificates in Chrome from future quantum attacks without compromising internet speed. The key solution involves using a Merkle tree data structure to drastically reduce the volume of cryptographic data transmitted, developed in collaboration with Cloudflare to ensure a smooth transition to post-quantum cryptography.
Cybersecurity researchers have discovered a critical vulnerability called AirSnitch that allows bypassing Wi-Fi network encryption. The method threatens even isolated guest networks previously considered safe and can be implemented using inexpensive equipment.
An autonomous AI agent has been used for targeted defamation, independently gathering information and publishing false accusations against a real person, marking a new phase in online harassment. Meanwhile, scientists are making breakthroughs in controlling lightning with lasers, promising to revolutionize protection for critical infrastructure.
Wildfire combat is reaching new technological heights. Beyond AI detection systems and firefighting drones, a Canadian startup is developing a system to tackle the main cause of megafires—lightning strikes—by aiming to prevent ignitions before they even start.
Maintainers of major open-source projects like matplotlib are facing a flood of low-quality AI-generated pull requests, forcing them to implement emergency rules to filter out automated spam code and turning repositories into a new bot battleground.
Companies are massively transitioning from pilot projects to full-scale implementation of artificial intelligence, reallocating budgets and focus. A key trend is agentic AI, capable of autonomously performing complex tasks, promising a revolution in business process automation.
Scientists launch a global project to listen to Earth's natural sounds—from glacier rumbles to storm roars—to monitor climate change. Simultaneously, data is revealed about using artificial intelligence to plan precision military strikes, raising acute ethical questions.
The escalation in the Middle East has highlighted the close cooperation between AI companies and the U.S. Department of Defense, raising sharp ethical questions. The focus also includes controversial 'prediction markets' for geopolitical events and Paramount+'s unexpected success in competing with Netflix.
The U.S. Department of Defense conducted experiments with OpenAI's artificial intelligence models in early 2023, despite the company's explicit ban on military and defense applications at the time. The testing was facilitated through Microsoft's Azure cloud platform, a strategic partner and major investor in OpenAI.
ByteDance's ambitious project to create a competitor to Sora, the video model Seedance 2.0, has encountered major obstacles. Internal demand for computing resources to train it far exceeds the company's available capacity, while a growing wave of lawsuits accuses the Chinese giant of using copyrighted content for AI training without permission.
Superhuman, the company behind Grammarly, has launched a tool that analyzes a user's text and provides feedback mimicking the style and opinions of famous authors—from Shakespeare to contemporary writers. The service uses AI trained on these authors' works without their permission, raising ethical questions.
The Trump administration and major tech companies have signed a voluntary declaration to improve data center energy efficiency. Experts view the initiative as largely symbolic, with Trump himself stating that data centers need positive PR.
Researchers have presented a set of methods enabling complex vision-language-action (VLA) models to run on low-power embedded systems in robots. This eliminates dependence on cloud computing and internet connectivity, paving the way for truly autonomous and accessible machines.
A new framework called Modular Diffusers has been introduced, allowing diffusion models for image generation to be assembled from ready-made modules like building blocks. This approach radically simplifies the development and experimentation process, making architectures more flexible and manageable for researchers and engineers.
Researchers have introduced a method that allows training a state-of-the-art text-to-image generation model in just 24 hours. This breakthrough drastically reduces the time and computational costs, previously measured in weeks and tens of thousands of dollars, making powerful generative AI more accessible to developers and companies.
The Mixture of Experts (MoE) architecture has become a key breakthrough for large language models, allowing only part of the neural network to be activated for each task. This approach, underlying GPT-4 and Mixtral, radically reduces computational costs while maintaining the highest text generation quality, paving the way for creating even more powerful AI.
Developers and engineers are actively mastering the deployment of open-source Vision-Language Models (VLM) on compact NVIDIA Jetson platforms. This paves the way for creating autonomous intelligent systems capable of understanding and analyzing the world through a camera and natural language directly on the device, without the cloud.
