Iran announced that data centers of giants like Google, Microsoft, and NVIDIA are now considered legitimate targets for military strikes. The Tasnim agency, linked to the IRGC, published a list of objects that could potentially be targeted in the future. For the first time, this list officially includes infrastructure of American tech companies in the Middle East: regional offices, server buildings, and research laboratories.
Iran attributes this to the connection between computing power and military and economic operations by the US and Israel. According to local authorities, current conflicts in the region are evolving into so-called infrastructure wars, and therefore the list of potential targets for cyber or physical attacks will continue to expand.
Meanwhile, Microsoft has defended Anthropic in its dispute with the Pentagon. In a San Francisco court, the corporation filed a motion to stop the US Department of Defense’s decision to classify Anthropic as a threat to supply chains. Microsoft requests a temporary halt to the blocking of existing contracts and to freeze relevant decisions.
The company believes that an immediate ban on using Anthropic’s technologies would force IT companies to radically change their architecture, potentially causing disruptions in existing AI systems used by the US military. A temporary delay would help avoid chaos and provide a smoother transition period for defense structures.
Microsoft’s main motivation is financial interest. Last November, it announced plans to invest up to $5 billion in Anthropic, while remaining a major investor in their competitor, OpenAI.
NVIDIA’s new Nemotron 3 Super model is an open development based on the MoE architecture with 12 billion active parameters and powered by 120V. It is primarily designed for autonomous agents tackling complex multi-step tasks. It employs a hybrid of Mamba layers—saving memory—and traditional transformers for high-level logical operations.
The model features an extended context window—up to one million tokens—and a multi-token generation function that triples processing speed. Nemotron 3 Super is available under an open license on the HuggingFace platform and has already been adopted by companies like Perplexity, Palantir, and Siemens.
Perplexity announced its own autonomous AI system based on Mac mini. The idea is to create an agent that operates 24/7 without human intervention: integrating with local files and applications on a dedicated device and autonomously performing complex tasks. The architecture combines local interaction with the desktop and heavy computational processes—handled by Perplexity’s servers.
The core of the system is a proprietary orchestration engine that automatically switches between different models to perform tasks. Users can manage this assistant remotely via any device. Security measures include emergency shutdown options, detailed session logs, and manual confirmation of sensitive commands. Enrollment in the project queue is available through a provided link.
Anthropic has launched a new research institute—Anthropic Institute. This division will focus on studying the global impact of AI on the economy, security, and society at large. The head is co-founder Jack Clark. Main areas of work include labor market transformation, risks of misuse of technology, and issues related to maintaining human control over autonomous AI systems.
The team comprises about 30 specialists from existing Anthropic departments, including the Frontier Red Team stress-testing group. Former researchers from organizations such as Google DeepMind and OpenAI have already joined.
Created with n8n:
https://cutt.ly/n8n
Created with syllaby:
https://cutt.ly/syllaby